Section 498A of the Indian Penal Code is the most argued-about provision in Indian criminal law, and the argument is conducted with two statistics that flatly contradict each other. One side says the conviction rate under the section is so low that most complaints must be false. The other side says the conviction rate is unremarkable for an offence committed inside a household and that the complaints are real. Both sides quote the National Crime Records Bureau. Both quotations are arithmetically correct, because the two percentages are computed on different denominators, and almost nobody who quotes either of them says which denominator it is. That single omission has sustained a public controversy for three decades, produced an extraordinary sequence of Supreme Court interventions in which the same court built a screening mechanism, demolished it as beyond its own function, and then reinstated it, and left the provision’s working almost untouched by Parliament throughout.

This article is the implementation and impact account of that provision. It states what the offence actually covers, reconstructs the record the enforcement system has produced, sets the two conviction figures side by side on their own denominators so that the reader can see what each one measures and what it cannot measure, narrates the judicial sequence with the precision it demands, states the misuse debate at full strength in the terms of the people who hold each side of it, and separates out the institutional question about the courts’ power to build screening machinery, which is a question about the separation of powers and deserves to be answered on its own terms rather than as a proxy in the argument about misuse. The article’s discipline throughout is the denominator discipline this series has applied to the enforcement records of the anti-terror statutes, the insolvency code and the money laundering statute: a percentage is an answer only when its base is named.

What Section 498A actually makes an offence

The provision was inserted into the Indian Penal Code by the Criminal Law (Second Amendment) Act, 1983. It provides that whoever, being the husband or a relative of the husband of a woman, subjects that woman to cruelty shall be punished with imprisonment for a term which may extend to three years, and shall also be liable to fine. The definition of cruelty sits inside the section, in an explanation with two limbs. The first limb covers wilful conduct which is of such a nature as is likely to drive the woman to commit suicide or to cause grave injury or danger to life, limb or health, whether mental or physical, of the woman. The second limb covers harassment of the woman where the harassment is with a view to coercing her or any person related to her to meet any unlawful demand for any property or valuable security, or is on account of failure by her or any person related to her to meet such a demand. The second limb is the dowry limb, and the statute whose demand offence it parallels, whose prohibition the penal provision was built to reinforce, is treated in the account of the Dowry Prohibition Act 1961, which owns that statute’s questions in this series.

Three features of the drafting carry the whole subsequent history. The offence is defined by relationship and by conduct, not by transaction: nothing need have been given, demanded successfully, or transferred, and the prosecution’s case is built from behaviour inside a household rather than from a paper trail. The class of possible accused is open-ended on the husband’s side, extending to any relative of the husband, with no requirement that the relative live in the same house, visit it, or have any particular proximity to the marriage. And the offence was classified from the beginning as cognizable, non-bailable and non-compoundable, the most coercive classification the procedural code ordinarily applies to an offence of this gravity, carrying a maximum sentence of three years. A qualification attaches to the cognizability: the First Schedule to the procedural code has always conditioned it on the information being given by the woman herself, by a relative of hers by blood, marriage or adoption, or, where no such relative exists, by a public servant of a class notified by the state government. The offence is therefore not one that a stranger or a neighbour can set in motion, a design feature that is easy to miss in the public argument and that matters when the argument turns to who files these complaints and why.

The offence is triable by a magistrate of the first class, and its sentence is discretionary up to the three-year maximum, with a fine that carries no statutory ceiling figure in the text. Sentencing practice in the field reflects the offence’s setting: where conviction follows a proved course of conduct unaccompanied by grave injury, courts have commonly imposed sentences well inside the maximum, and the appellate moderation recorded later in this article, reducing sentences after settlement to the period already undergone, presupposes that pattern. The three-year ceiling also fixes the provision’s procedural neighbourhood: it places the offence inside the class punishable with up to seven years to which the arrest safeguards of 2014 were addressed, and inside the ordinary magistrate trial system whose throughput the next sections measure, rather than in the sessions courts that try the cluster’s graver offences of dowry death and abetment.

The provision did not arrive alone. The same amending Act of 1983 inserted Section 113A into the Indian Evidence Act, under which, where a married woman commits suicide within seven years of her marriage and it is shown that her husband or a relative of her husband subjected her to cruelty, the court may presume, having regard to all the other circumstances of the case, that the suicide was abetted by the husband or the relative. The presumption is discretionary, where the dowry death presumption inserted three years later is mandatory, a distinction the evidence articles in this series have been careful to preserve. The cruelty offence and the suicide presumption were a pair: the first punished the conduct, the second adjusted the burden of proof for the conduct’s gravest outcome, in recognition that both occur in a setting where outside witnesses are rare and the household’s own members are the accused’s family.

The recodification of the criminal law in 2023 carried the offence into the new codes without diluting it. In the comparison of the Penal Code with the Bharatiya Nyaya Sanhita, the mapping is set out in full: the offence stands at Section 85 of the Bharatiya Nyaya Sanhita, 2023, and the definition of cruelty, reproducing the two limbs, stands at Section 86. The punishment is unchanged at three years and fine. The classification is unchanged: cognizable on the same qualified information, non-bailable, non-compoundable. Conduct before the new code commenced continues to be charged under the Penal Code provision, so both citations will run in parallel in the courts for years, and every safeguard the courts developed under the older numbering applies to the newer one, because the text it interprets is materially the same text.

The procedure that governs arrest under the successor provision sits in the Bharatiya Nagarik Suraksha Sanhita, 2023, whose Section 35 gathers the arrest powers formerly spread across Sections 41 and 41A of the Code of Criminal Procedure. Its scheme is the statutory home of the safeguard jurisprudence this article narrates: for an offence punishable with imprisonment of up to seven years, arrest is not a default but a decision, requiring the officer to be satisfied that arrest is necessary for one of the stated purposes, such as preventing further offences, securing proper investigation, or preventing the person from tampering with evidence or absconding, and to record the reasons in writing; and where arrest is not required, the officer is to issue a notice directing the person to appear, with appearance and continued compliance answering the state’s need without custody. The Supreme Court’s 2014 directions, examined below, were built on the predecessor provisions, and their substance survives the renumbering because the successor text preserves the structure they enforced.

Why was the offence made cognizable and non-bailable in 1983?

The classification was a response to the enforcement failure of the law that preceded it. Dowry demands and cruelty inside the matrimonial home had been addressable, if at all, through a prohibition statute that prosecuted transactions and exposed the giving family to prosecution alongside the taking one, and through ordinary hurt and intimidation offences that treated each incident in isolation.

A cognizable and non-bailable cruelty offence let the state act on the woman’s complaint without a magistrate’s prior permission and without the accused’s release being a matter of right, and a non-compoundable offence prevented the complaint from being traded away inside the family under the same pressure that produced it.

The pressure that produced the provision is part of the dowry cluster’s history, and this article compresses it because the cluster’s pillar owns it. A joint committee of Parliament that examined the working of the Dowry Prohibition Act reported in 1982 and refused to treat that Act’s failure as a failure of drafting detail; it recommended reaching beyond the prohibition to the criminal code itself, on the reasoning that a practice enforced by cruelty inside the household had to be met by offences addressed to the cruelty and to its fatal outcomes. The amendments of 1983 and 1986 followed that design: definition and prohibition in the Dowry Prohibition Act, conduct and death in the penal code, and evidence presumptions calibrated to each in the Evidence Act. Section 498A was the conduct offence, the one designed for daily use, and its subsequent history is the history of what daily use did to it.

The offence’s edges: husband, wife, relative and the threshold of cruelty

The provision’s open texture at its edges has generated its own interpretive record, and three edges matter to the implementation account because each determines who enters the criminal process at all. The first edge is the word husband and the phrase relative of the husband. In U. Suvetha v. State by Inspector of Police, decided in 2009, the Supreme Court held that a relative for this purpose must be a person related to the husband by blood, marriage or adoption, and that the husband’s girlfriend or concubine, however central her role in the marriage’s breakdown, is not a relative of the husband within the section and cannot be charged under it. The holding draws the accused class’s outer boundary at legal relationship, which matters in practice because complaints not infrequently name a woman alleged to be involved with the husband, and such names must fall away at the first judicial scrutiny. Inside the boundary, the class remains wide: parents, siblings, and relatives by marriage are all within it, with no requirement of shared residence, which is why the specificity jurisprudence, rather than the definition, became the operative filter on the class’s width.

The second edge is the word wife. The offence protects a woman subjected to cruelty by her husband, and the question whether the protection extends to a woman whose marriage is void or voidable, or whose relationship is a marriage in form but not in law, was answered expansively in Reema Aggarwal v. Anupam, decided in 2004, where the Court held that a woman who had gone through a ceremony of marriage with a man already married, and was treated by him as his wife, could invoke the provision, reasoning that the section’s purpose would be defeated if a man could escape it by pleading the invalidity of the very marriage he had contracted. The courts have since applied the protective provisions to relationships in the nature of marriage in cognate contexts, and the practical position is that the provision looks to the substance of the marital relationship rather than to the husband’s technical defences about its validity.

The third edge is the threshold of the first limb itself. Not every unhappy marriage is cruel within the section, and the case law has consistently required conduct of gravity: wilful conduct of such a nature as is likely to drive the woman to suicide or to cause grave injury or danger to life, limb or health. Two disciplines operate on this threshold. The first is that the conduct is judged as a course of conduct where the complaint presents one, so that acts individually small may accumulate into the statutory gravity, which is the complainant-side reading the courts accept. The second, operating in the opposite direction, is that the ordinary wear and tear of matrimonial life, disagreements over money, household arrangements, food, tempers and the friction of a joint household, do not cross the threshold merely because they are narrated in a complaint, and quashing courts have repeatedly struck out proceedings whose allegations, taken at their highest, describe marital unhappiness rather than the defined cruelty. The tension between those two disciplines is not resolvable in the abstract; it is the daily work of the quashing jurisdiction described in this article, and it is one reason the quality of a complaint’s pleading has come to matter so much: the same household history can be pleaded as a course of coercive conduct or as a catalogue of grievances, and the courts’ threshold jurisprudence rewards the first and penalises the second.

The presumption beside the offence: Section 113A at work

The cruelty offence’s gravest cases are those in which the woman dies, and the evidence presumption enacted beside it is the instrument the system built for exactly those cases, so its working belongs in this account. Where a married woman commits suicide within seven years of her marriage, and the prosecution shows that her husband or a relative of her husband subjected her to cruelty within the section’s meaning, Section 113A of the Evidence Act, carried into Section 117 of the Bharatiya Sakshya Adhiniyam, permits the court to presume that the suicide was abetted by the husband or the relative. The presumption’s architecture is deliberate at every joint. It does not arise from the suicide alone: the prosecution must first prove the cruelty, to the criminal standard, as the foundation on which the presumption is raised. It is discretionary in terms, the court may presume, having regard to all the other circumstances of the case, so a court that finds the foundation proved may still decline the inference where the surrounding circumstances point the other way. And it operates on abetment, the charge under the suicide provision that ordinarily requires proof of instigation or intentional aid, substituting an inference drawn from the proved cruelty and the marriage’s recency for direct proof of the instigating act, in recognition that the instigation of a suicide inside a household is, of all facts, the least likely to be witnessed.

The presumption’s pairing with the offence explains a charging pattern visible throughout the record: the cruelty charge and the abetment charge travel together wherever a marriage has ended in suicide within the period, and the cruelty evidence does double duty, founding both the cruelty conviction and the presumption that supports the graver one. The calibration against the neighbouring presumption is the point the evidence articles in this series have insisted on and this article repeats because the public debate habitually confuses the two: for dowry death, the presumption enacted in 1986 is mandatory, the court shall presume, once the foundational facts of a dowry-linked death within seven years are shown; for abetment of suicide, the presumption is discretionary, the court may presume, and the difference between those two verbs is a considered legislative judgment about how much of the prosecution’s burden each setting can bear. In the implementation record the discretionary presumption has functioned as its design suggests: a bridge available in the strongest cruelty cases rather than a substitute for proof, invoked where the proved course of conduct makes the inference natural and declined where the marriage’s history, though unhappy, does not carry the inference to the grave.

The machinery the provision required, and what was actually built

An implementation account begins with the machinery. Section 498A required no new institution, no new cadre, no budget line and no rules to be framed before it could operate, and that is the first fact about its implementation: unlike statutes whose promises wait on machinery that is never built, this provision ran on the ordinary criminal justice system from the day it commenced. Its implementation architecture is the ordinary architecture, a police station that registers the complaint, an investigation, a charge sheet, a magistrate’s court for trial, and the remand and bail jurisdiction that surrounds arrest. Everything that has happened to the provision since 1983 has happened inside that ordinary machinery, which means its failures and its excesses are the system’s failures and excesses wearing a particular section number.

The complaint enters the system in one of two ways. The woman, or a relative of hers within the statutory classes, may give information to the police, who register a first information report; or she may file a complaint before a magistrate, who may direct investigation or take cognizance. The Supreme Court’s constitution bench decision in Lalita Kumari v. State of Uttar Pradesh, delivered in 2013, placed matrimonial and family disputes among the categories in which a preliminary inquiry may be conducted before a first information report is registered, to ascertain whether a cognizable offence is disclosed at all. That qualification, little known outside the profession, is the first screening device the system applies to these complaints, and it predates every committee experiment this article describes. Around the police station there also grew, unevenly across states, an informal layer of women’s cells, counselling units and legal services clinics to which matrimonial complaints are frequently routed first, a layer with no statutory basis in the provision itself and very variable practice, which the formal record barely captures.

Investigation follows the ordinary pattern, statements of the woman and her family, statements of neighbours and of the husband’s household where they will give them, medical records where injury is alleged, and the documentary residue of the marriage’s finances where a demand is alleged. The investigation’s structural difficulty is written into the offence: the conduct occurs in private, the witnesses are overwhelmingly members of one or other family, and the events complained of are usually a course of conduct over months or years rather than a single dated incident, so the charge sheet that emerges tends to narrate a pattern in general terms, a feature that returns at the quashing stage and again at trial. The Bureau’s figures record that the overwhelming majority of investigated cruelty cases end in a charge sheet rather than in a final report closing the case, a proportion the Supreme Court put at 93.6 per cent when it quoted the Bureau’s 2012 data in Arnesh Kumar, and a figure each side of the debate reads in opposite senses, as the debate section below sets out.

Arrest is where the machinery’s design and its practice diverged most sharply, and the divergence is the hinge of the whole judicial story. The statutory design, even before 2014, never made arrest automatic: the arrest power for an offence of this gravity was governed by the necessity conditions in the procedural code, and the notice of appearance stood in the code from 2009 as the alternative where arrest was not necessary. The practice, as the Supreme Court found in 2014, was that arrest followed registration as a matter of routine, that the conditions were not applied, that reasons were not recorded, and that magistrates authorised detention mechanically on remand applications. The distance between the code’s text and the station’s practice, rather than anything in the cruelty provision itself, is what the arrest jurisprudence of the following decade was built to close, and it is why that jurisprudence is addressed to police officers and magistrates rather than to the offence.

How does a cruelty complaint move through the system?

A complaint by the woman or a qualifying relative is registered, sometimes after a preliminary inquiry because the dispute is matrimonial; the police investigate a course of conduct that usually has no neutral or independent witnesses; most investigated cases are charge-sheeted.

The accused are then arrested or, under the safeguard regime, noticed to appear; a magistrate tries the case over years in a queue of pending trials; and the case ends, in proportions the next section measures, in conviction, in acquittal, in a quashing on settlement, or in the large residue still pending.

The trial end of the machinery carries the system’s general burden in an acute form. Cruelty is the single largest head of crime against women in the Bureau’s classification, which means the largest single stream of cases entering magistrates’ courts under that chapter, and the courts’ throughput is governed by the same constraints that govern every head: judicial vacancies, the availability of witnesses years after the event, and the priority that sessions trials for graver offences properly claim. The Bureau’s court disposal tables, examined in the next section, record pendency percentages for this head above ninety per cent, which is to say that of the total stock of cruelty cases standing for trial in a year, the courts dispose of fewer than one in ten. A provision can be perfectly drafted and still produce, at that throughput, a process whose duration is itself the principal sanction and the principal grievance, for complainants and accused alike. Much of what is attributed to the section’s design, on both sides of the debate, is attributable to this arithmetic.

Bail and the accused’s path through the safeguard regime

Because the offence is non-bailable, the accused’s first legal problem is liberty, and the path to it runs through three doors whose relative importance the safeguard regime rearranged. The first door is the notice of appearance: where the police officer, applying the necessity conditions, decides arrest is not required, the accused is not arrested at all, and the criminal process proceeds on appearance, which is the outcome the Arnesh Kumar directions were designed to make ordinary in cases of this gravity. The second door is anticipatory bail, the High Court’s or sessions court’s power, placed in Section 482 of the Bharatiya Nagarik Suraksha Sanhita, to direct release in the event of arrest; it was in an anticipatory bail matter that Arnesh Kumar itself arose, and applications of this kind remain a standard early move for accused relatives, frequently met with interim protection while the application is heard, which converts the arrest question into a largely paper contest conducted before custody occurs. The third door is regular bail after arrest, where the magistrate’s obligation to record satisfaction before authorising detention, and the Satender Kumar Antil consequence that arrest in breach of the safeguards supports release, have changed the remand hearing from a formality into a genuine checkpoint in the districts where the directions are observed.

The regime’s incompleteness is as instructive as its design. The directions bind police officers and magistrates, and their sanctions, departmental action and contempt, are real but rarely invoked; compliance is therefore highest where High Courts have actively supervised it and most variable in the districts furthest from that supervision, so that the practical answer to whether arrest follows a cruelty complaint depends on geography in a way the law’s text nowhere admits. Accused persons with counsel learn the regime and use it; accused persons without counsel experience the older system the directions were meant to displace, which is the same distributional pattern the quashing jurisdiction shows. And the complainant’s experience of the same regime is the mirror image: a notice issued instead of an arrest may be the system working exactly as directed, and may also be the moment at which pressure on her to settle becomes irresistible, a duality the defenders’ case presses and the directions themselves do not answer. The safeguard regime, on the whole record, is best understood as a genuine and measurable reform of the process’s first step that leaves the process’s remaining length, cost and attrition exactly where they were, which is why the debate it was meant to quiet continues at undiminished volume around it.

The numbers, and the denominator discipline

The series to which this article belongs has applied one method to every enforcement record it has examined: before quoting any percentage, name its denominator, its source and its period, and refuse to compare percentages computed on different bases. The method dissolved the enforcement argument in the anti-terror cluster, where the enforcement record of the anti-terror statutes showed how a conviction figure changes meaning with its base; it did the same work in the implementation record of the Insolvency and Bankruptcy Code, where admission and resolution figures are routinely divided by the wrong totals; and again in the account of the money laundering statute’s Supreme Court judgment, where attachment values and prosecution numbers circulate without their bases. This is the fifth application of that discipline, and the most politically charged, which is precisely why the method matters: the same arithmetic that dissolved the argument in those clusters dissolves it here, and it does so without favouring either side, because what it dissolves is not a conclusion but a confusion.

Start with the figures that are not in dispute. The Bureau recorded 1,40,019 cases of cruelty by a husband or his relatives registered in 2022, the largest single head of crime against women in its classification, at 31.4 per cent of all cases registered under that chapter. Registrations had risen from 1,13,403 in 2015. Over the same span, the Bureau’s figures record the number of persons arrested under the head falling from 1,87,067 in 2015 to 1,45,095 in 2022. The two series moving in opposite directions is the statistical signature of the Arnesh Kumar regime: more complaints entering the system, fewer arrests per complaint, with the arrest decision increasingly taken by notice rather than by custody. Persons arrested exceed cases registered in both years, which is the arithmetic shadow of the provision’s open-ended accused class: the average cruelty case names more than one accused.

The trend across the years is worth stating as a series, because single-year figures in this debate are usually chosen for effect. The Bureau recorded 1,03,272 cruelty cases in 2018, 1,25,298 in 2019, 1,11,549 in 2020, the year in which lockdown conditions suppressed registration across most heads of crime, 1,36,234 in 2021 and 1,40,019 in 2022, with 1,33,676 recorded in 2023. Read whole, the series shows a large and broadly stable registration volume with growth across the period, not an explosion and not a decline, and it shows the pandemic year’s dip and recovery that every registration-based series in the Bureau’s volumes displays. Neither side’s rhetoric maps cleanly onto that shape: the volume is too large and too persistent to be an artefact of any single litigation fashion, and too stable to support claims of a sudden epidemic of falsehood, while remaining, on the survey evidence, a minority of the underlying conduct.

Geography complicates every national figure in this field, and the Bureau’s own caution, printed under its state tables, that states may not be compared purely on the basis of crime figures applies with special force here. The largest absolute numbers of crimes against women are recorded in Uttar Pradesh, and the cruelty head’s largest state counts are recorded in West Bengal and Uttar Pradesh, while recorded rates per lakh of female population vary across states by factors that no serious reader attributes to equivalent variation in marital conduct. Registration is a behaviour before it is a statistic: it depends on the distance to a police station, the station’s practice with matrimonial complaints, the presence of women’s cells and legal aid, the complainant’s natal family’s support, and the local profession’s habits, all of which vary more than the underlying phenomenon plausibly does. A state with a high recorded cruelty rate may have more cruelty, more willingness to record it, or both, and the Bureau’s tables cannot separate those. The practical consequence for the debate is that state-level conviction comparisons, which circulate freely in it, inherit the same defect as the national percentages in an aggravated form, since they divide an outcome shaped by court capacity by a registration shaped by police practice and call the quotient a measure of complainants.

At the court end, the Bureau’s Crime in India 2022 reports a conviction rate of 17.7 per cent for this head in its court disposal tables. The Bureau’s conviction rate has a fixed meaning in its tables: it is the number of cases in which trial ended in conviction, divided by the number of cases in which trial was completed in the year, where a completed trial is one that ended in conviction or acquittal on the merits. It therefore excludes every case that ended without a trial, the cases quashed, the cases compounded or compromised, the cases withdrawn or abated, and the great mass of cases still pending, and it excludes them by design, because it is a measure of what trial courts do with the trials they finish. The figure for 2019, the last pre-pandemic year the Bureau published in the same format before the series this article draws on, was 14.6 per cent, on 40,785 completed trials. A rate in the region of a sixth to a fifth, on completed trials, is the first of the two headline figures.

The second headline figure is computed by taking the same numerator, the cases ending in conviction, and dividing it by the total stock of cases standing for trial in the year, which is the cases carried over from previous years plus the cases sent for trial in the year. For 2019 the Bureau’s tables record 51,396 cruelty cases disposed of by the courts in the year and 5,70,695 still pending trial at its end, a total stock of 6,22,091 cases, with a pendency percentage of 91.7 per cent. The cases ending in conviction in that year, applying the Bureau’s published rate of 14.6 per cent to its 40,785 completed trials, number about 5,950. Divided by the total stock, that is just under one per cent. The figure is arithmetically correct. It is also a measure of something entirely different from the first figure: it measures the courts’ throughput against their backlog as much as it measures outcomes, because more than nine tenths of its denominator is cases that have not been decided at all, including cases that will in time be quashed, compromised, acquitted and convicted in proportions nobody yet knows.

There is an irony in the provenance of the two figures that an implementation account should not pass over. The modern form of the argument was set by the Supreme Court itself, in Arnesh Kumar, where the judgment placed two of the Bureau’s rates side by side in a single passage: a charge-sheeting rate of 93.6 per cent, computed on investigated cases, and a conviction rate in the region of fifteen per cent, computed on completed trials, quoted from the Bureau’s 2012 data. Each figure was accurate, each base was different, and the judgment’s rhetorical movement from the first to the second invited the reader to feel the drop between them as a measure of the complaints’ quality, although the drop is mostly the distance between two denominators with the entire court process in between. The decade of public argument that followed largely consisted of the two halves of that passage being quoted separately by the two sides, the defenders keeping the charge-sheeting rate and the critics keeping the conviction rate, each discarding the base the other had kept. The denominator discipline exists, as a method, because the field’s founding document committed the field’s founding error.

Between those two poles sit the intermediate bases, and the table below sets them out, because the table is the object this argument has always needed and never had. Each row is a denominator in public circulation. The conviction count is held constant at the 2019 figure of about 5,950 cases, computed from the Bureau’s published rate and completed trials for that year, so that the reader can watch a single numerator change its meaning as the base beneath it changes. The 2022 figures are given in the rows where they belong, with their own year attached, and are not blended into the 2019 arithmetic.

Denominator	Figure, with source and period	Convictions as a share of this base	Who quotes this base	What it does and does not measure
Cases registered in the year	1,25,298 cases (NCRB, Crime in India 2019); 1,40,019 in 2022	About 4.8 per cent (2019 convictions over 2019 registrations)	Rarely quoted directly; the base both sides gesture at when speaking of complaints	Measures complaints entering the system against convictions emerging from it in the same calendar year, although the convictions arise overwhelmingly from earlier years’ complaints; it does not measure the truth of either year’s complaints
Cases charge-sheeted after investigation	About 93.6 per cent of investigated cases charge-sheeted (NCRB 2012 data, as quoted by the Supreme Court in Arnesh Kumar)	Not the base of either headline figure; the convictions that follow charge sheets are counted in the court rows below	Defenders quote the rate to show complaints survive professional screening; critics quote it to show the screening is nominal	Measures the police decision to send a case to court; a charge sheet records a prima facie view on the complainant’s material, not a finding, and a high rate is consistent with both diligence and deference
Cases disposed of by courts in the year	51,396 cases (NCRB, Crime in India 2019, court disposal tables)	About 11.6 per cent	Court administrators and some researchers	Measures convictions against everything the courts closed in the year, including cases quashed, compounded and withdrawn without any finding on the merits; it does not isolate the trials the courts actually decided
Trials completed on the merits	40,785 cases (NCRB, Crime in India 2019); the Bureau’s conviction rate for 2022 on this base is 17.7 per cent	14.6 per cent in 2019; 17.7 per cent in 2022; this is the Bureau’s published conviction rate	The Bureau itself, and defenders of the provision answering the misuse claim	Measures the outcome of trials that ran their course; it says nothing about cases that never reached a verdict, and a low rate on this base is consistent with weak investigation, hostile witnesses, settlement pressure and false complaints in unknown proportions
Total cases standing for trial, including the pending stock and cases later quashed or compromised	6,22,091 cases (51,396 disposed plus 5,70,695 pending at the end of 2019, NCRB)	Just under 1 per cent	Critics of the provision, as the headline proof that almost no complaint ends in conviction	Measures convictions against a stock more than nine tenths of which is undecided; it is principally a measure of pendency and throughput, and it cannot distinguish a false complaint from a true one still waiting for trial
Cases ending in conviction	About 5,950 cases (computed from the Bureau’s published 14.6 per cent rate applied to its 40,785 completed trials, 2019)	The numerator held constant across every row	Both sides, always without the base attached	The only figure in the table that is an outcome; every dispute in this field is a dispute about what to divide it by

What a genuine misuse rate would require is worth specifying, because specifying it shows why the debate cannot be settled by citation and what would settle it. A misuse rate is a proportion of complaints that were false when made. No existing series counts it, but a measurement is constructible: a coded sample of closed cruelty cases, drawn across states and years, in which trained readers record from the file itself the terminal event of each case, conviction, acquittal on the merits after a contested trial, acquittal following the complainant’s evidence collapsing or turning hostile, discharge, quashing on the merits, quashing on settlement, closure by the police as false with the magistrate’s acceptance, and withdrawal, together with the closing court’s recorded reasons where reasons exist. Such a study would not produce the single number the debate wants, because falsity is a finding files rarely record in terms, but it would replace the present contest of denominators with a distribution of endings, in which settlement, proof failure and judicially recorded incredibility each had a measured share. Every participant in this debate who is confident of the answer should be demanding that study; that none of the organised participants has produced it, in three decades, while quoting the Bureau’s tables at each other, is the clearest available evidence that the debate’s function for its participants is not wholly evidentiary.

Three lessons follow from the table, and they are lessons about method rather than about the provision. First, neither headline figure is a misuse rate. A misuse rate would require knowing how many complaints were false, and no Indian statistical series counts that: the Bureau counts outcomes, and an acquittal records that an offence was not proved to the criminal standard, not that it was invented, while a case closed by the police as false is recorded at the investigation stage on the officer’s assessment and is itself contested. Secondly, the two figures answer different questions, and each is the right answer to its own question. Asked how a finished trial under this head usually ends, the answer is conviction in roughly one case in six or seven. Asked what fraction of the mountain of cases standing before the courts a year’s convictions represent, the answer is under one in a hundred, which is a statement about the mountain. Thirdly, any speaker who quotes a conviction percentage in this debate without naming the base, the source and the year has told the audience nothing, and the audience is entitled to treat the omission as the argument.

What do the arrest figures show since the safeguard regime began?

They show a system registering more complaints and arresting less readily. Registrations of cruelty cases rose from 1,13,403 in 2015 to 1,40,019 in 2022 on the Bureau’s figures, while persons arrested under the head fell from 1,87,067 to 1,45,095 over the same span.

The divergence dates from the 2014 arrest directions and is the clearest quantitative evidence that the directions changed behaviour at the police station, whatever view is taken of the provision itself.

The judicial sequence: a mechanism created, struck down and reinstated

The provision’s implementation history is, in its decisive chapters, a history of the Supreme Court managing the consequences of a statute it did not write and could not amend. The sequence has five acts in the Court, one in a High Court, and a sixth act in the Court again, and it must be narrated in order because each act is unintelligible without the one before it.

The first act is validation with a warning. In Sushil Kumar Sharma v. Union of India, decided in 2005, the Supreme Court upheld the constitutional validity of Section 498A against a challenge that the provision was being misused and was capable of striking terror in innocent families. The Court held that the possibility of misuse does not invalidate a provision enacted for a legitimate object, that the object here, striking at the dowry menace and at cruelty within marriage, was plainly legitimate, and that the remedy for misuse lay in the ordinary disciplines of investigation, trial and appellate correction rather than in striking the provision down. In the same judgment the Court recorded the concern in language that has been quoted in every subsequent round of the debate: the provision was intended to be used as a shield and not as a weapon, and by misuse of the provision a new legal terrorism could be unleashed. The Court commended the concern to the legislature. The judgment thus settled the constitutional question while formally opening the implementation question, and it is the implementation question that the Court itself then took up.

Between validation and intervention, the Court’s concern sharpened in a series of observations in quashing and bail matters. In Preeti Gupta v. State of Jharkhand, decided in 2010, the Court recorded a noticeable tendency to implicate the husband and all his immediate relations in complaints under the section, observed that a large number of such complaints were not bona fide and sprang from trivial matrimonial discord, exhorted courts to take care, and again addressed the legislature, observing that a serious relook at the entire provision was warranted. These observations matter to the sequence because they show the Court moving from recording misuse in 2005 to characterising its scale by 2010, while still directing its remedy at Parliament. Parliament did not act. The Court then acted itself.

The second act is Arnesh Kumar v. State of Bihar, decided on 2 July 2014, and it is the most consequential single judgment in the provision’s working life. The case arose from a cruelty complaint and an application for anticipatory bail, and the Court used it to address arrest practice across the board. It recorded that arrest in these cases had become routine, that the provision’s cognizable and non-bailable character had placed it among the provisions used as weapons rather than shields by disgruntled wives, in the judgment’s phrase, and it quoted the Bureau’s data for 2012, a charge-sheeting rate of 93.6 per cent and a conviction rate in the region of fifteen per cent, alongside the arrest figures showing how many of those arrested were the husband’s relatives, including elderly parents and women. The operative directions were addressed to the machinery, and they applied not to cruelty cases alone but to all offences punishable with imprisonment of up to seven years. A police officer was not to arrest automatically on registration; the officer was to satisfy himself against the checklist of necessity conditions in Section 41 of the Code of Criminal Procedure that arrest was required, to record the reasons for arresting and for not arresting, and to forward the checklist and reasons to the magistrate within the period governing production and remand. The notice of appearance under Section 41A was to be the instrument where arrest was not necessary. A magistrate authorising detention was to do so only on recorded satisfaction from the material forwarded, not mechanically. And the directions carried sanctions: an officer who arrested without complying was exposed to departmental action and to contempt proceedings before the High Court, and a magistrate who authorised detention without recording satisfaction was exposed to departmental action by the High Court. A later bench in Satender Kumar Antil v. Central Bureau of Investigation, decided in 2022, reaffirmed the Arnesh Kumar discipline and attached the practical consequence that non-compliance supports release on bail, completing the conversion of the directions from exhortation into a working rule with a remedy.

The third act went a step further, and the step was the one that broke. In Rajesh Sharma v. State of Uttar Pradesh, decided on 27 July 2017, a two-judge bench directed the constitution of Family Welfare Committees in every district, composed of paralegal volunteers, social workers, retired persons and other suitable persons, to be constituted by the District Legal Services Authority. Every complaint under the section received at a police station or before a magistrate was to be referred to the committee; the committee could interact with the parties and their family members and was to prepare a report within a month; and no arrest was ordinarily to be made until the committee’s report was received, with an exception where the complaint disclosed tangible physical injury or death. The bench issued further directions about settlement, the clubbing of connected proceedings, and the conduct of trials, and it grounded the experiment in the same data and the same concern as its predecessors. The distinction from Arnesh Kumar was categorical, and everyone in the field saw it at once: Arnesh Kumar had enforced the procedural code’s own arrest conditions against the officers who were ignoring them, while Rajesh Sharma created a body the code did not contain and interposed it, with a suspensive effect on arrest, between the complainant and the process the statute prescribed.

The fourth act was the correction. The women’s organisations intervened, the All India Democratic Women’s Association among them, arguing that the committees would be extrajudicial bodies of lay persons, akin to community panchayats, standing between a woman and the criminal justice system, and that the conviction figures on which the experiment rested did not show what they were said to show. In Social Action Forum for Manav Adhikar v. Union of India, decided on 14 September 2018, a three-judge bench presided over by the Chief Justice held that the direction to constitute Family Welfare Committees was impermissible: the creation of such a mechanism was not within the judicial function, it amounted to legislation by the court in an area the statute had occupied, and the directions were set aside to that extent. The bench was careful about what it preserved. The Arnesh Kumar safeguards were retained as the governing discipline on arrest, being enforcement of the code rather than addition to it. The settlement route was confirmed: where parties settle, the course is an application to the High Court under its inherent powers, applying the principles of Gian Singh. And the legislative question, whether the offence should be made compoundable or its procedure redesigned, was returned to Parliament, where the Court said it belonged. The institutional holding was as clear as a holding can be: a court administering a statute may discipline the actors the statute appoints, but it may not appoint new actors of its own.

The fifth act took place in a High Court, and it is the turn in the sequence that makes the story institutionally strange. In Mukesh Bansal v. State of Uttar Pradesh, by a judgment of 13 June 2022, the Allahabad High Court framed its own guidelines for cruelty cases, expressly drawing guidance, as it recorded, from the Supreme Court’s 2018 judgment. The guidelines provided that after the lodging of a first information report or complaint, no arrest or coercive police action should be taken until a cooling period of two months had concluded; that during that period the matter was to be referred to a Family Welfare Committee in each district, constituted under the District Legal Services Authority with at least three members, whose constitution and functioning were to be reviewed periodically by the District and Sessions Judge; and that the referral applied to cruelty cases accompanied by other offences punishable with imprisonment of less than ten years, with cases involving injury of the gravest kind treated separately. A High Court had therefore rebuilt, within its own territorial jurisdiction, substantially the mechanism that the Supreme Court had held four years earlier to be beyond the judicial function. The guidelines operated in Uttar Pradesh, the state that registers the largest volume of crimes against women in the Bureau’s tables, and they operated in a space the 2018 judgment appeared to have closed.

The sixth act is Shivangi Bansal v. Sahib Bansal, decided on 22 July 2025 by a bench presided over by the Chief Justice. The case reached the Court as a transfer petition arising out of the parties’ own litigation, a marriage that had generated a thicket of criminal, domestic violence, maintenance and civil proceedings against each other and each other’s families in more than one state, in the course of which the husband had spent 109 days in custody and his father 103 days. The Court, exercising its power under Article 142 of the Constitution to do complete justice, dissolved the marriage, quashed the proceedings between the parties, and recorded that the wife would issue an unconditional public apology to the husband and his family. It then addressed the Allahabad guidelines. It approved them. It directed that the safeguards framed by the Allahabad High Court regarding the constitution of Family Welfare Committees to prevent misuse of the provision should be followed by the authorities and would remain operational. The mechanism that a three-judge bench had set aside in 2018 as an impermissible act of judicial legislation was thereby reinstated with the Supreme Court’s authority, adopted from a High Court’s judgment rather than created by the Court’s own directions, and carried into force through the Court’s power to do complete justice in the case before it rather than through a reconsideration of the 2018 holding, which the judgment did not purport to overrule.

That is the sequence, and its strangeness should be stated plainly because the article’s readers include the journalists and researchers who will have to explain it. The central procedural question in this field, whether complaints under the cruelty provision should be screened by a lay committee before the criminal process moves, has been answered three different ways by the same institution in eight years: directed in 2017, forbidden as beyond the judicial function in 2018, and approved in 2025. At no point in the sequence did Parliament amend the provision, and at no point did the statutory text under interpretation change in any respect material to the question. The statute stood still while the institution’s account of its own powers moved. Commentators across the debate’s divide have identified the resulting inconsistency, and it is a live institutional question, examined on its own terms in a later section of this article: not whether screening is wise, on which reasonable positions differ, but by what route a mechanism held to be beyond judicial creation in 2018 became operative national practice in 2025 without the 2018 holding being revisited by a bench of co-equal or greater strength.

What the reinstated mechanism actually requires

Because the 2025 approval gives the Allahabad guidelines operative force, their practical content belongs in this account, stated as the guidelines state it rather than as either side’s summary of it. The mechanism applies at the threshold of the criminal process. On the lodging of a first information report or a complaint under the cruelty provision, the matter is to be referred at once to the Family Welfare Committee for the district, and for a cooling period of two months from the lodging, no arrest is to be made and no coercive police action taken against the named accused. The referral is confined to cruelty cases, including complaints in which the cruelty provision is accompanied by other offences punishable with imprisonment of less than ten years; complaints disclosing injury of the gravest character stand outside the scheme. Each district is to have at least one committee, constituted under the District Legal Services Authority, with at least three members, and its constitution and functioning are to be reviewed periodically by the District and Sessions Judge or the Principal Judge of the family court, who chairs or co-chairs the district legal services authority. The committee’s work during the cooling period is of the kind its name describes: it receives the complaint, interacts with the parties and their families, and attempts the reconciliation or settlement that the criminal process itself has no organ for attempting, reporting back into a process that then proceeds, if it proceeds, with the committee’s view of the dispute on the file.

Three features of the design deserve the attention of any reader assessing it, whatever view they take of the institutional question. The first is that the committee’s intervention is temporal rather than adjudicatory: it decides nothing, convicts nobody and acquits nobody; its power is the power of delay plus the power of a mediated conversation at the moment of maximum heat, and its defenders and critics are really arguing about the value and the danger of that delay. The second is that the design places the committee inside the legal services architecture rather than inside the police, which answers part of the 2018 objection about lay bodies of unexamined competence, since the District Legal Services Authority is a statutory body chaired by the district judiciary, although it does not answer the objection that the committee itself, its membership and its procedure, are creatures of guidelines rather than of any enacted text. The third is that a two-month suspensive period interacts with everything else in the complainant’s position: evidence is not perishable in a course-of-conduct offence in the way it is in a stranger assault, which is the design’s best factual support, but the complainant’s safety, residence and economic position during those two months are exactly what the civil statute’s interim orders exist to secure, which means the mechanism’s safety depends on the parallel machinery working in the districts where it is thinnest. The mechanism as designed is neither the nullification its fiercest critics describe nor the cure its framers hoped; it is a structured pause, and the record of its operation in Uttar Pradesh since 2022, on which any final assessment must rest, has not been published in a form that permits one.

The quashing line: the courts’ second answer

Running beneath the sequence about committees is a second, quieter body of doctrine that does more daily work than any committee: the High Courts’ power to quash. Because the offence is non-compoundable, a cruelty prosecution cannot be ended by the parties’ agreement in the trial court, however complete the settlement and however genuine the reconciliation or the parting. The parties’ only exit, short of trial, is the High Court’s inherent power, preserved in Section 482 of the Code of Criminal Procedure and carried into Section 528 of the Bharatiya Nagarik Suraksha Sanhita, to quash proceedings to prevent abuse of the process of the court or to secure the ends of justice. The doctrinal line that governs its use in matrimonial cases was built in three decisions and refined in a continuing series of specificity cases.

In B.S. Joshi v. State of Haryana, decided in 2003, the Supreme Court held that the bar on compounding a non-compoundable offence does not bar the High Court from quashing such proceedings where the parties have settled, because the object of the inherent power is different from the object of the compounding provisions: compounding is the parties’ act recorded by the court, while quashing on settlement is the court’s own judgment that continuing the prosecution would serve no purpose and would itself be an abuse. In Gian Singh v. State of Punjab, decided in 2012 by a three-judge bench, the Court systematised the position: offences of a private character, including matrimonial offences, may be quashed on genuine settlement notwithstanding their classification, while offences with a public character, economic offences affecting the financial system and offences under special statutes stand on a different footing. The cruelty offence sits squarely in the private class, and the consequence is that one of the most common endings of a cruelty case, visible in the Bureau’s tables in the compounded and compromised column and in the quashing column, is a negotiated ending ratified by a High Court, an ending the statute’s text nowhere provides for and the courts have supplied because the alternative was trials nobody wanted at the end of marriages that had already ended.

The second branch of the quashing line requires no settlement at all. It descends from the categories in State of Haryana v. Bhajan Lal, decided in 1992, under which proceedings may be quashed where the allegations, taken at their face value, do not disclose the offence, or are so absurd or inherently improbable that no prudent person could proceed on them. Applied to the cruelty provision, this branch has generated the specificity jurisprudence, a line of decisions insisting that allegations against each accused be specific as to act, occasion and role, and that omnibus allegations against the husband’s relatives, of the kind that narrate a household’s unhappiness in general terms and attribute it to everyone in it, do not make out the offence against those relatives. In Geeta Mehrotra v. State of Uttar Pradesh, decided in 2012, the Court quashed against relatives where the complaint made only casual reference to them without specific allegations of their conduct. In Kahkashan Kausar v. State of Bihar, decided in 2022, the Court held that general and omnibus allegations against the husband’s relatives cannot be permitted to rope them into trial, warning against the tendency to implicate extended family on vague material. And in Dara Lakshmi Narayana v. State of Telangana, decided in December 2024, the Court quashed proceedings against six appellants, observing that it had uncovered an ongoing tendency to use the provision as an instrument of personal vendetta against the husband and his family, and that courts must scrutinise such complaints with care. The line is not indulgent to complainants or to accused as classes; it is a discipline about pleading, and its practical content is that the quality of the complaint’s drafting determines, to a degree unusual in criminal law, how much of a cruelty case survives its first judicial examination.

The quashing line matters to the implementation account for three reasons. It is the safety valve that the non-compoundable design required and the statute omitted, and its existence is the courts’ own admission, repeated across two decades, that a prosecution the parties have settled should not be kept alive by classification alone. It is also the principal filter the system applies to the over-inclusion problem, operating accused by accused, which is a finer instrument than any committee screening a complaint as a whole. And it distributes its benefits unevenly, because reaching a High Court with a quashing petition requires counsel, fees and time, so the filter works best for accused families with resources, a distributional fact that belongs in any honest assessment of the regime the courts have built around this provision.

Which other provisions travel with a cruelty complaint?

A cruelty complaint rarely travels alone. The same facts support charges under the Dowry Prohibition Act for giving, taking or demanding dowry, a charge of criminal breach of trust over the woman’s own property where it is withheld, and, where the marriage has ended in death or suicide, the graver provisions on dowry death and abetment with their calibrated presumptions.

The cluster is charged together, defended together, and very often settled together, which is why the provision’s statistics cannot be read as the record of a single offence standing alone.

Two adjacent doctrines complete the practical picture and are persistently misunderstood. The first concerns place. In Rupali Devi v. State of Uttar Pradesh, decided in 2019, the Supreme Court held that a woman who leaves the matrimonial home and takes shelter at her parental home may file the complaint there, because the cruelty’s consequences continue at the place of shelter and the courts there have jurisdiction; the complaint is not confined to the matrimonial home’s territory. The second concerns time. The limitation period for an offence punishable with up to three years is three years under the procedural code, but the courts have treated cruelty as a continuing offence whose clock runs from the course of conduct rather than from its first incident, and have condoned delay where the conduct is shown to have continued, with the result that limitation arguments in this field succeed rarely and on particular facts rather than as a general defence. Neither doctrine appears in the provision’s text, and both are part of its working law.

One front in a wider war: the cruelty case among the matrimonial cases

No account of the provision’s working is complete without placing the criminal case inside the litigation system in which it actually lives. A marriage’s breakdown in India rarely produces one case. It produces a constellation: the criminal complaint under the cruelty provision and its satellite offences; proceedings under the Domestic Violence Act for protection, residence and monetary relief; a maintenance claim; divorce proceedings on grounds that commonly include cruelty in its matrimonial sense; custody and guardianship contests; and frequently a criminal breach of trust claim over the woman’s property and proceedings under the Dowry Prohibition Act. The Shivangi Bansal litigation, in which proceedings between one couple and their families ran across two states and every one of these categories, is an unusually well documented specimen of an ordinary structure. Each proceeding has its own forum, its own timeline and its own interim orders, and each is capable of being used, by either side, as pressure in the negotiation that will eventually settle all of them at once.

This structure explains features of the record that single-case analysis cannot. It explains why settlements, when they come, are global, and why the criminal case is so often the last proceeding closed rather than the first, since it is the one whose closure requires a High Court’s indulgence and therefore the one with the greatest exchange value in the negotiation. It explains the timing of complaints, which cluster at the breakdown’s crisis points, after separation, after a divorce petition, after a maintenance order, in patterns that each side reads tendentiously: the critics read the sequence as evidence that the complaint is a counter-move in litigation, the defenders read it as the entirely predictable consequence of the fact that a woman complains of cruelty when the marriage ends, since complaining while it continues carries costs the breakdown has already imposed. Both readings fit the same timestamps, and the record does not distinguish them, which is a further instance of the article’s central methodological point. And the structure explains why the criminal provision, among all the instruments in the constellation, attracts the controversy: it is the only one that arrests. Maintenance arrears do not put a father-in-law in a police station; a residence order does not require a married sister in another city to seek anticipatory bail. The state’s coercive apparatus enters the family’s dispute through this provision alone, and the argument about the provision is, at bottom, an argument about whether that entry, on those terms, is the protection the legislature promised in 1983 or a hazard the courts have spent forty years learning to fence.

The practice layer: how the provision is actually operated

Between the statute and the statistics sits a layer of professional practice that no judgment prescribes and no report measures, and an implementation account that omitted it would describe a system nobody uses. The first practitioner is the drafter of the complaint. Because the offence is a course-of-conduct offence and the quashing jurisprudence rewards specificity, the complaint in a contested case is usually a drafted document, prepared with legal assistance, narrating the marriage from the wedding’s expenses through the demands alleged, the departures, the returns and the final rupture, and naming the accused in descending order of proximity. The drafting convention of the field, narration first and section numbers last, means the document reads as a family history with a criminal conclusion, and both the police officer registering it and the magistrate reading it later are responding to a narrative whose selection and emphasis were chosen adversarially. The second practitioner is the station house officer, whose real decision, beneath the formal one about arrest, is whether the complaint is routed to the counselling and women’s cell layer first or registered at once, a routing decision governed by local practice, by the complaint’s presentation, and by the parties’ relative insistence, and one that determines whether the case enters the Bureau’s statistics this month, next month, or not at all.

The third practitioner is the mediator, because the system’s settlement work is increasingly done in court-annexed mediation centres and in legal services authority processes attached to the matrimonial litigation, where the criminal case is one item on a settlement agenda that includes the divorce, the maintenance figure, the custody arrangement and the return of property. The settlement that results is characteristically a package with a single consideration: peace. Its terms commonly include the quashing petition the High Court will be asked to allow, the mutual withdrawal of the constellation’s other proceedings, and a lump sum or structured maintenance in full settlement, which is why the criminal case’s closure so often trails the civil settlement by exactly the time a quashing petition takes. The fourth practitioner is the trial advocate, whose working knowledge includes the field’s open secrets: that the complainant’s evidence will be tested against her statements in the parallel proceedings, that inconsistencies between the criminal complaint, the domestic violence application and the divorce petition are the defence’s principal material, and that the prosecution’s principal material is the consistency of the core account across the same three documents. A field in which the same facts must be sworn to in four forums generates its own failure rate through inconsistency alone, independent of any question of truth, and that generated failure is part of what the acquittal figures contain and cannot show.

The misuse debate, stated at full strength on both sides

The debate is about what the record assembled above means, and it is conducted by organised participants who have held their positions, with evidence genuinely available to each, for three decades. This article states each case in its holders’ own terms before assessing any of it, and it does not characterise either side’s motives, because neither side’s case depends on the other’s bad faith and both are weakened when their opponents pretend otherwise.

The critics’ case begins from the provision’s structure. The offence is cognizable and non-bailable; the accused class is any relative of the husband; a single complaint therefore reaches parents, married sisters living in other cities, and elderly relatives at the moment of a marriage’s collapse, and arrest, in the decades before the safeguard regime, followed registration as a matter of course. The consequences of arrest are not undone by acquittal: employment is lost, reputations in the professions and in government service are marked, elderly accused are put through remand and bail proceedings, and the family’s standing in its community is damaged in ways no later order repairs. The critics then turn to outcomes. Conviction, they observe, is the terminus of a small minority of complaints on any base wider than completed trials, and a process that ends in conviction so rarely must, in their reading, be starting from complaints that should never have been made in large numbers. The non-compoundable classification completes the design complaint: parties who have reconciled, or who have settled their affairs and wish only to end the litigation, cannot end it, and the criminal case survives as leverage in the negotiation over maintenance, custody and property long after its ostensible subject has been resolved. The institutional holders of this case are named and on the record. The Save Indian Family Foundation and allied men’s rights organisations have campaigned on it for decades. The Supreme Court’s own benches supplied its canonical statements in Sushil Kumar Sharma, in Preeti Gupta and in Arnesh Kumar. The Malimath Committee on reforms of the criminal justice system, reporting in 2003, recommended that the offence be made bailable and compoundable. The Committee of Petitions of the Rajya Sabha, in a report presented on 7 September 2011, recommended that the government consider making it compoundable. The Law Commission, in its 243rd Report of 2012 devoted to the provision, recommended that it be made compoundable with the permission of the court and subject to a cooling-off period of three months, while recommending that it remain non-bailable and that the answer to arbitrary arrest lie in strict observance of the arrest provisions. Parliamentary standing committees had recommended compoundability in 2005 and again in 2008. The critics’ centre of gravity, across these holders, is not repeal: it is the conversion of a non-compoundable offence into a settleable one, coupled with discipline on arrest, and the record of their case is that governments of both national coalitions have declined to enact even that, through a complete recodification of the criminal law in which the offence was carried forward unchanged.

The defenders’ case begins from the baseline the critics’ figures omit. Cruelty in the matrimonial home is pervasive and registered only in fragments: the National Family Health Survey’s fifth round, covering 2019 to 2021, found that 29.3 per cent of ever-married women aged eighteen to forty-nine had experienced spousal violence, a prevalence measured in tens of millions of women, against which the Bureau’s lakh and a half of annual registrations is a small fraction; and of the women who experienced violence, only about fourteen per cent sought help from any source. On that baseline, the defenders argue, rising registration is at least as consistent with rising access and awareness as with rising falsity, and any account of misuse that does not begin from the unregistered majority is measuring the visible tip and pronouncing on the whole. They then answer the outcome figures directly. A low conviction rate, in their reading, reflects the offence’s evidentiary structure rather than its complainants’ honesty: the conduct occurs in private, the witnesses are the accused’s family or the complainant’s, complainants withdraw under family and economic pressure during trials that run for years, witnesses turn hostile, settlements are reached and then prosecuted half-heartedly, and the standard of proof is the criminal standard applied to a course of conduct rather than to a single witnessed act. An acquittal establishes that the offence was not proved; it does not establish, and methodologically cannot establish, that the complaint was false, and the Bureau’s statistics are not compiled in a form that can measure misuse at all, a point made in scholarly form by Ruchira Goswami of the West Bengal National University of Juridical Sciences. The defenders quote the screening figure in the opposite sense from the critics: more than ninety per cent of investigated cases are charge-sheeted after police investigation, which the All India Democratic Women’s Association, intervening against the Family Welfare Committee directions, read as evidence of the pervasive nature of the violence rather than of its invention, adding that cases fail for want of proper investigation, omitted witnesses and compromise under the pressure of delay. Against screening committees specifically, the defenders’ case is at its strongest: a filter that delays police action for two months operates in exactly the period when a woman who has finally complained is at greatest risk of retaliation and of pressure to withdraw, and it interposes lay persons of unexamined competence between the complainant and the system at the moment of maximum vulnerability. Flavia Agnes, the women’s rights lawyer associated with the Majlis legal centre in Mumbai, made the geographical version of the argument in print as early as 1999: that it is easy, sitting in a metropolis, to pass judgment that the laws are being misused, when the larger reality is of laws that have yet to reach minimum standards of use. Kirti Singh, a former member of the Law Commission, has opposed dilution as counterproductive and pointed to the complaints never registered or never followed up as the system’s larger error.

Certain facts are common ground, and identifying them narrows the genuine dispute. Both sides accept that automatic arrest on registration was a documented abuse and that the Arnesh Kumar discipline was justified; the arrest figures since 2015 show the discipline working, and no significant holder on the defenders’ side campaigns for a return to automatic arrest. Both sides accept that omnibus implication of relatives on vague allegations occurs and is wrong; the specificity jurisprudence is the defenders’ inheritance as much as the critics’, since a complaint that fails for vagueness damages the genuine complaints that follow it. Both sides accept that the process is too slow and that pendency near or above ninety per cent is a scandal independent of anyone’s position on the offence. The genuine residue of dispute is threefold. It is, first, a values dispute about where the risk of error should fall: on the accused family exposed to a process triggered by complaint, or on the complainant exposed to delay and filtration at the point of danger, and evidence cannot settle that distribution. It is, secondly, a factual dispute about the composition of the non-conviction majority, how much of it is fabrication, how much is proof failure, and how much is settlement, on which the available statistics are structurally silent, as the denominator table shows. And it is, thirdly, a dispute about compoundability that is narrower than the rhetoric on either side suggests, since the institutional centre, the Law Commission’s court-supervised compounding after a cooling-off period, has been on the table since 2012, is opposed by some defenders as a pressure channel and supported by many critics as a settlement route, and has been declined by every government that has faced it, which is itself evidence that the political cost of being seen to dilute the provision exceeds, for those who must decide, the institutional consensus in its favour.

A further strand of the critics’ case, held most consistently by the organised men’s rights movement, should be recorded because it proposes a different reform from all the others: that the offence is gendered in its very definition, protecting a wife against a husband and his relatives while offering no corresponding protection to a husband against cruelty by a wife and her relatives, and that protective criminal legislation of this one-sided character offends equality. The answer on the record is constitutional and empirical in combination: Article 15(3) of the Constitution expressly permits special provision for women, and the offence’s defenders point to the asymmetry of the underlying phenomenon as the survey data record it, including the dowry economy that the provision’s demand limb targets and that has no mirror image on the other side. Parliament, in the recodification, retained the gendered definition without alteration, and no court has read the provision as gender-neutral or struck it for one-sidedness; the challenge in Sushil Kumar Sharma included the provision’s alleged capacity for misuse and was answered on the footing that protective classification of this kind is the legislature’s to make. The strand matters to the debate’s future because it is unreconciled rather than answered in the political arena: its holders regard every procedural safeguard as a partial correction of a substantive wrong, which is why safeguard settlements that satisfy the compoundability centre do not satisfy them.

Individual cases periodically reset the debate’s temperature, and the mechanism deserves recording because it has become a structural feature of the field. In December 2024 the death by suicide of a Bengaluru software professional, Atul Subhash, who left a long note and a video attributing his decision in substantial part to the matrimonial litigation pending against him and describing the proceedings from the accused’s side in detail, convulsed public discussion of the provision for weeks and carried the critics’ case to an audience no report reaches. The case is part of this article’s record for what it did rather than for what it proved: the allegations in the note were the account of a participant, the proceedings it described were contested, and a death of that kind is evidence of the process’s human cost and of the despair litigation can produce, not an adjudication of any complaint’s truth. But its effect on the debate was real and asymmetric, because the defenders’ case, being statistical and structural, has no equivalent artefact, and a field whose public understanding is periodically reset by individual tragedies, on whichever side they occur, will keep misunderstanding the aggregate record this article has set out. The Supreme Court’s own cautionary observations in a quashing appeal decided the same month, in Dara Lakshmi Narayana, show the institution registering the same weather. An implementation account cannot insulate the provision from that cycle; it can only insist, as this article has, that the aggregate record and the individual case answer different questions, and that policy made from either alone will be wrong.

The institutional question, on its own terms

The separation of powers question raised by the sequence deserves treatment separate from the misuse debate, because it would matter even if every complaint under the provision were genuine and every arrest justified, and even if the opposite were true. The question is this: may a court, administering a statute, create a screening body that the statute does not contain and suspend the statute’s ordinary process pending that body’s report?

The case for the negative is the 2018 bench’s case, and it is a strong one. The criminal process is a statutory scheme in which each actor’s role, the police officer who registers and investigates, the magistrate who takes cognizance and remands, the prosecutor who conducts the trial, is assigned by the code. A committee of paralegal volunteers and social workers, screening complaints and reporting before arrest, is a new actor with a suspensive power over the scheme’s first coercive step. Creating it is legislation in the strict sense: it adds an institution, a procedure and a delay that the enacted text does not authorise, and it does so for a class of cases defined by the court. The objection is reinforced by the design’s own features: the committee’s members are appointed by no statutory process, its procedure is prescribed by no rules, its report’s legal status is undefined, and the complainant whose case it screens has no appeal from its effective consequences. On this view, the vice of the 2017 directions was not their policy, which many who opposed the directions shared in part, but their form: a change of this kind is exactly what the legislative process, with its committee scrutiny and its accountability, exists to make, and a court that makes it teaches every future bench that the boundary is optional.

The case for the affirmative is weaker in doctrine and stronger in circumstance, and it should also be stated fairly. The courts’ power to issue directions to fill a vacuum in the law’s working is an established feature of Indian constitutional practice, exercised most famously in the guidelines jurisprudence of the 1990s, and its defenders point out that the arrest safeguards of 2014, which nobody of consequence disputes, were themselves directions issued in a case, refining a code that Parliament had already amended in the same direction in 2008 without securing compliance. On this view, the distinction between enforcing the code’s arrest conditions and adding a screening step is one of degree along a single spectrum of judicial management of a failing process, and the 2018 line was drawn at a point the Court’s own practice does not obviously compel. The defenders of the 2025 outcome add a further point about the route: the Court in Shivangi Bansal did not create the mechanism but approved guidelines framed by a High Court within its supervisory jurisdiction over the courts and processes of its state, and it acted under Article 142, the power to pass such orders as are necessary for doing complete justice in a cause before it, a power the Constitution’s text confers in the widest terms.

What can be said neutrally, and what the record supports, is narrower than either case and more uncomfortable than both. The 2018 holding has not been overruled, distinguished by a larger bench, or reconsidered; it stands as the Court’s stated account of the limits of its function, delivered by three judges after argument in which the committee experiment was the very question. The 2025 approval, delivered by two judges in a transfer petition whose lis was the parties’ own litigation, gives the contrary practice the Court’s authority without engaging that account. The two positions therefore coexist without a reconciling principle that a district judge, a police officer or a legal services authority can extract and apply: the mechanism is impermissible for a Supreme Court bench to create and permissible for it to bless, and the difference lies in the route rather than in any stated principle about the mechanism itself. Institutional questions of this kind do not stay academic. The legal services authorities that must constitute the committees, the magistrates who must decide whether to act before a report arrives, and the complainants and accused whose cases turn on the answer, all require a rule, and the sequence has given them an outcome in search of one. Whether the resolution comes from a larger bench rationalising the two positions or from Parliament occupying the field with a designed procedure, the field’s central procedural question remains, on the Court’s own record, answered inconsistently. That is the namable claim this article carries: the same mechanism, created, struck down and reinstated, with the statute unchanged throughout.

Why does the separation of powers question matter here?

Because the answer determines who may redesign the criminal process. If screening committees are legislation, only Parliament can create them, and the 2025 position cannot be generalised beyond the case that produced it. If they are judicial management of a statutory process, every High Court may build a variant, and the procedure in a cruelty case will differ by state.

The Supreme Court has held the first in terms and done the second in practice.

The parallel civil route and the choices a complainant actually has

The criminal provision is not the only instrument addressed to cruelty in the matrimonial home, and an implementation account that ignored the parallel route would misdescribe the system complainants actually navigate. The Protection of Women from Domestic Violence Act, 2005 created a civil track running alongside the criminal one: on a complaint of domestic violence, which includes the same cruelty in civil dress, a magistrate may make protection orders restraining further violence, residence orders securing the woman’s place in the shared household, monetary reliefs for expenses and losses, compensation orders, and interim and ex parte orders at speed, supported by Protection Officers and registered service providers. The full account of that statute, its definitions as read by the courts and its reliefs in operation, is given in the article on the Domestic Violence Act 2005, and this article does no more than mark the relationship. The civil route requires no arrest, names no accused in a criminal sense, moves faster, and can protect residence and income while a marriage is sorted out; the criminal route carries the state’s coercive weight and the stigma of prosecution, and it is the route whose pendency and process this article has measured. The two routes are not alternatives in law, since both may be pursued together, but they are alternatives in strategy, and a substantial share of complainants, on the accounts of practitioners in this field, do not know the civil route exists when they first approach the system, an information failure that costs nothing to repair and is repaired nowhere systematically.

The practical residue of the record, for the reader who arrives with a problem rather than a research question, can be stated in the terms the preceding sections have established. Arrest on a cruelty complaint is not automatic in law and has not been since 2014: the officer must record reasons against the statutory checklist or issue a notice of appearance, and a magistrate must record satisfaction before authorising detention, with consequences for default on both sides of that transaction; an accused person who receives a notice should treat appearance and cooperation as the substance of the safeguard, since non-compliance with a notice is itself a ground for arrest. A complaint that names relatives must still survive the specificity discipline, and a complainant’s case is strengthened by dates, occasions and identified acts rather than by the length of the list of the accused. Parties who have settled have a real remedy despite the non-compoundable classification, in a quashing petition to the High Court on the Gian Singh principles, and parties who have not settled should expect the trial to be measured in years, which is an argument for the civil route’s interim protections being sought in parallel rather than after. The general position described in this article is legal information about a provision and its record; it is not advice on any individual case, and forum, limitation and the facts of a particular marriage govern outcomes in ways no general account can anticipate. A reader preparing the figures in this article for teaching or presentation can open the slide deck in a browser-based PowerPoint viewer to check the denominator table’s presentation before relying on it, which is the use the table was built for.

The trial, and why so few cases end in a verdict at all

The denominator table records that of more than six lakh cruelty cases standing for trial in 2019, about forty thousand trials were completed. The anatomy of the uncompleted remainder is the least examined part of the implementation record, and it explains more of the conviction argument than either side’s account of the complainants or the police. A cruelty trial is built, in the ordinary case, on the testimony of the complainant, supported by her parents and siblings, by such neighbours as will speak, by medical records where treatment was sought, and by the documentary traces of demands where demands were reduced to writing or to messages. Against it stands the accused household’s denial and, frequently, its own witnesses from the same closed setting. The offence’s subject matter guarantees this structure: cruelty is a course of conduct in a private space, and there is no forensic substrate, no scene, no instrument, no neutral record of the ordinary kind. The trial court is therefore asked, years later, to reconstruct a marriage’s interior from the accounts of two families who are, by the time of trial, usually also litigating divorce, maintenance, custody and property against each other, each with every incentive the adversarial process supplies.

Three consequences follow, and each is visible in the outcome pattern. The first is the hostile witness phenomenon: complainants and their relatives resile from their statements in large numbers, whether because a settlement has been reached whose terms the criminal case would disturb, because the complainant has rebuilt her life and will not spend it in a witness box, because family pressure has done its work over the years of pendency, or because the original statement was embellished and cannot be sustained. The prosecution, deprived of its witnesses, ends in acquittal, and the acquittal enters the statistics as a datum in the misuse argument although its causes are mostly unrecorded and mostly unknowable from the file. The second consequence is the mid-trial settlement: a large share of cases that do not end in verdict end because the parties settled everything, the matrimonial litigation, the property, the custody and the criminal case in one negotiation, and the criminal case was then closed by quashing on the settlement or allowed to wither into a compromise recorded by the court, a category the Bureau’s tables carry in the tens of thousands across the years, recording about fifteen and a half thousand cruelty cases as compounded or compromised in 2022 notwithstanding the offence’s formal non-compoundability. The third consequence is appellate moderation: where conviction does occur and the parties thereafter settle, appellate courts have in suitable cases reduced the sentence to the period already undergone, treating the settlement as a reason to close the matter’s penal chapter, so that even the conviction column contains cases whose practical end was negotiation.

None of this is an argument that the trials that do complete are wrongly decided, in either direction. It is an argument about what a conviction rate built on completed trials can and cannot carry. A rate of conviction in one completed trial in six or seven, in a field where the evidence is family testimony about private conduct tested years later at the criminal standard, is not self-evidently a scandal, and it sits within the range the Bureau records for other offences of private setting and relational evidence. What the figure cannot do, and what both sides ask it to do, is distribute the non-convictions among fabrication, proof failure and settlement. The Bureau does not ask courts why cases failed. Until some statistical series does, the misuse debate’s central factual question remains unanswerable from the public record, and every participant who claims the record answers it is overclaiming.

How the figures travel: the public life of two percentages

The discourse around this provision, in print, on broadcast panels and in organised campaigning, is where the denominator error does its practical damage, and an implementation account should record the mechanisms of the damage because they are part of the provision’s impact. The first mechanism is the bare percentage: a conviction figure quoted with no base, no source and no year, which hardens by repetition into a fact about the provision itself. The second is the silent base-switch: an argument that begins with the sub-one-per-cent figure computed on the total stock of cases, to establish that conviction is vanishingly rare, and continues as though it had established that complaints are almost always false, which is a different proposition resting on a base that includes lakhs of undecided cases. The third is the mirror image on the other side: an argument that begins with the charge-sheeting rate, to establish that complaints almost always survive investigation, and continues as though investigation were a finding of truth, when a charge sheet records only that the police found material fit to place before a court. The fourth is the cohort error: dividing one year’s convictions by the same year’s registrations, although the convictions arise from complaints registered years earlier, an error that flatters or damns the system depending on whether registrations are rising or falling, and that the table in this article commits openly in one row precisely so that the reader can see it labelled.

The errors are not distributed symmetrically in their consequences. The critics’ headline figure, honestly computed, measures pendency and calls it falsity, and its circulation has built the political weather in which screening mechanisms of doubtful legality appeared necessary and were cheered. The defenders’ headline figures, honestly computed, measure prevalence and screening and call them vindication, and their circulation has built the counter-weather in which any procedural reform, including reforms the defenders’ own institutional holders have supported, is treated as dilution and resisted. A journalist covering this field, who is one of this article’s intended readers, has exactly one professional obligation that resolves most of the difficulty: never print a percentage in this debate without its base, its source and its year in the same sentence. The obligation costs a line of type. Its absence has cost the field three decades of argument at cross purposes.

The comparative check: an unused law and a machinery-bound law

Template discipline in this series requires an implementation account to check its subject against a comparable statute, and two comparisons discipline the argument here. The first is with the Dowry Prohibition Act itself, the provision’s parent statute in the same cluster. Nobody campaigns about the misuse of the Dowry Prohibition Act, for the reason the pillar article in this series identifies: the Act is barely used, its symmetrical offence structure deters the very complaints it exists to receive, and its annual registrations run at roughly a tenth of the cruelty provision’s. The misuse debate, in other words, is a debate about a working law. A provision that nobody invokes generates no arrest grievances, no quashing jurisprudence and no organised opposition; it also protects nobody. The comparison disciplines the critics’ implicit baseline: the alternative to a heavily used and contested enforcement provision, in this field’s own history, is not a lightly used and harmonious one but an unused one, and the dowry cluster contains both specimens side by side for any reader who wishes to compare their social products.

The second comparison is with the Domestic Violence Act, the civil sibling, whose implementation problem is the opposite of the cruelty provision’s. The civil statute depends on machinery, Protection Officers in every district, registered service providers, magistrates’ capacity for urgent interim orders, and its monitoring reports over the years have dwelt on the thinness of that machinery, officers holding the charge in addition to other duties, service providers unevenly registered, orders delayed. The cruelty provision depends on no machinery beyond the police station and the magistrate’s court, and its implementation problem is not absence but indiscipline: actors present in full number, applying a discretionary power by default instead of by judgment. The two statutes thus exhibit the two classic failure modes of Indian social legislation in a single household’s case: the remedy that waits for machinery that was never staffed, and the remedy whose machinery works exactly as built except at the one point, the arrest decision, where judgment was required and routine took its place. It is not a coincidence that the judicial response concentrated on that point, nor that the civil route’s weakness is the one no judgment can repair, because constituting and funding a protection service is Parliament’s and the executive’s work in a way that disciplining arrest never was.

The reform menu, and what has actually been chosen

The proposals for reforming the provision’s working have been stable for two decades, and listing them against the record of what was chosen is the shortest way to see where the system’s real decisions were taken. The first proposal is compoundability with safeguards: the Law Commission’s 2012 package, compounding permitted by the court after a cooling-off period of three months, the offence remaining cognizable and non-bailable, arrest disciplined by the procedural code. The second is the Malimath package, broader and older: the offence made bailable as well as compoundable, treating the matrimonial dispute as a matter to be ended by settlement wherever settlement is genuine. The third is screening by design: a statutory preliminary process, whether police inquiry of the Lalita Kumari kind given structure and time limits, or a committee created by legislation with appointed members, defined powers and an appeal, doing openly what the courts’ committees attempted. The fourth is evidentiary and statistical reform: requiring the Bureau’s tables, or a successor series, to record the reasons cases end without conviction in a sample of courts, so that the debate’s central unknown becomes knowable. The fifth is throughput reform, more magistrates and faster trials, which commands universal verbal assent and addresses the pendency figure that powers the critics’ headline percentage.

Against that menu, the record of choice is instructive. The arrest discipline was chosen, by the courts, in 2014, and it worked measurably. Compoundability was chosen by nobody: recommended by a committee of Parliament in 2005, by another in 2008, by the Rajya Sabha’s Committee of Petitions in 2011 and by the Law Commission in 2012, it was omitted from the amendment legislation of 2013 that recast the sexual offence provisions of the same code in the same Parliament, and omitted again from the recodification of 2023, which re-enacted the offence in substance exactly. Screening was chosen three times in three incompatible forms, by two courts and by endorsement, without legislation. Statistical reform has never been chosen by anyone with the power to choose it. Throughput reform is chosen annually in speeches. The pattern is not one of a system that examined the menu and picked the best items; it is a pattern in which the one reform the courts could make alone was made by the courts, the reforms requiring legislation were made by no one, and the political economy of the subject, in which any dilution is a headline and any retention is a headline of the opposite kind, rewarded the legislature for leaving the provision exactly as it stood while its working was redesigned around it by judges. A reader who understands that pattern understands the provision’s implementation better than a reader who knows every case in the sequence, because the pattern predicts the future: absent a government willing to spend the political capital, the next change in this field will also arrive by judgment, and will also be attacked, from whichever side it disappoints, as legislation in robes.

The verdict the record supports

The honest assessment of the provision’s implementation separates three propositions that the public argument habitually fuses. The first is about the offence’s necessity, and on it the record is one-sided: the underlying conduct is prevalent on survey evidence at a scale registrations understate by an order of magnitude, the conduct was effectively unreachable by the law that preceded the provision, and no serious institutional holder proposes the provision’s repeal. The second proposition is about the process the provision triggers, and on it the record is also one-sided, in the other direction: the process was for decades operated with automatic arrest in defiance of the procedural code’s own text, it runs at a throughput that leaves more than nine in ten cases undecided at each year’s end, and its duration inflicts serious and irreversible costs on complainants and accused alike, costs that fall hardest on those without the resources to reach a High Court’s quashing jurisdiction. The third proposition is about misuse, and on it the record is genuinely incomplete in both directions: misuse occurs, as every bench that has examined the question has found and as the specificity jurisprudence presupposes, and its scale is not measurable from the statistics either side quotes, because acquittal cannot show falsity, charge-sheeting cannot show truth, and the only honest sentence the figures support is that the proven residue of the debate is smaller than both sides’ rhetoric and larger than either side’s concessions.

The judicial response, taken as a whole, divides into the part that enforced the law and the part that attempted to improve it. The enforcement part, the arrest discipline of Arnesh Kumar preserved in 2018 and reaffirmed since, is the courts’ unambiguous success in this field: it changed measurable behaviour, it did so by holding the system’s own actors to the system’s own text, and it commands assent across the debate. The specificity discipline in the quashing cases belongs to the same family, a court doing the recognisably judicial work of insisting that allegations be allegations. The improvement part, the committee experiment, is where the institution overreached in 2017, corrected itself in 2018, and unsettled its own correction in 2025 without explaining the reconciliation, and the cost of that unsettlement is borne not by the Court but by every district functionary who must operate a mechanism whose constitutional warrant the Court has both denied and conferred. Parliament, which alone could have designed the screening question openly, with evidence, committee scrutiny and a text that magistrates could apply, declined the question at every opportunity across four decades, including in a recodification that re-enacted the offence word for word in substance. The provision’s implementation record is therefore best read as a record of institutional substitution: where the legislature would not act, the courts acted in its place, first well within their function and then beyond it, and the distance between those two kinds of action is the true subject of the controversy that the misuse debate, for all its volume, has largely served to obscure.

Frequently Asked Questions

Q: What does Section 498A make an offence?

Section 498A of the Indian Penal Code makes it an offence for a husband, or a relative of a husband, to subject a woman to cruelty, punishable with imprisonment of up to three years and a fine. Cruelty is defined inside the section in two limbs: wilful conduct of such a nature as is likely to drive the woman to suicide or to cause grave injury or danger to her life, limb or health, whether mental or physical; and harassment of the woman with a view to coercing her or any person related to her to meet an unlawful demand for property or valuable security, or on account of failure to meet such a demand. The offence is cognizable on the information of the woman or her close relatives, non-bailable and non-compoundable.

Q: Where does Section 498A sit in the new criminal code?

The offence was carried into the Bharatiya Nyaya Sanhita, 2023 without dilution. The offence itself stands at Section 85 of the Sanhita, and the definition of cruelty, reproducing the two limbs of the older provision, wilful conduct likely to drive the woman to suicide or cause grave injury or danger, and harassment to coerce an unlawful demand for property or valuable security, stands at Section 86. The punishment remains imprisonment of up to three years and a fine, and the classification remains cognizable on qualified information, non-bailable and non-compoundable. Conduct committed before the new code commenced continues to be charged under the Penal Code provision, so both citations run in parallel, and the judicial safeguards developed under the older numbering apply to the newer one because the operative text is materially unchanged.

Q: What did the Arnesh Kumar judgment direct about arrest?

In Arnesh Kumar v. State of Bihar, decided in 2014, the Supreme Court directed that arrest under Section 498A, and under every offence punishable with imprisonment of up to seven years, must not be automatic on the registration of a complaint. The police officer must satisfy himself against the checklist of necessity conditions in Section 41 of the Code of Criminal Procedure that arrest is required, record the reasons for arresting and for not arresting, and forward the checklist to the magistrate; where arrest is not required, the officer is to issue the notice of appearance under Section 41A. A magistrate authorising detention must record satisfaction from the material forwarded rather than acting mechanically. Officers who arrest in breach face departmental action and contempt; magistrates face departmental action by the High Court.

Q: Why were Family Welfare Committees struck down and then restored?

In Rajesh Sharma in 2017 the Supreme Court directed that every cruelty complaint be referred to a district Family Welfare Committee and that no arrest ordinarily be made until the committee reported. In Social Action Forum for Manav Adhikar in 2018, a three-judge bench set those directions aside, holding that creating such committees was not within the judicial function and amounted to legislation in an area the statute occupied, while preserving the Arnesh Kumar arrest safeguards. The mechanism returned by a different route: the Allahabad High Court framed its own guidelines in Mukesh Bansal in 2022, with a two-month cooling period and committee referral, and in Shivangi Bansal in 2025 the Supreme Court approved those guidelines and directed that they be followed, acting under its power to do complete justice, without overruling the 2018 holding.

Q: What does the conviction rate data actually show?

It shows two different things, depending on the denominator. The National Crime Records Bureau’s published conviction rate divides convictions by trials completed on the merits: 14.6 per cent for 2019 and 17.7 per cent for 2022, so roughly one completed trial in six or seven ends in conviction. Dividing the same convictions by the total stock of cases standing for trial, which for 2019 was 6,22,091 cases including 5,70,695 still pending, produces a figure just under one per cent, which measures pendency and court throughput more than outcomes. Neither figure is a misuse rate: an acquittal records that an offence was not proved to the criminal standard, and no Bureau series counts how many complaints were false when made.

Q: Can a cruelty case be settled or quashed?

Not by compounding, because the offence is non-compoundable and the trial court cannot record the parties’ compromise as a compounding. The working route is quashing: the High Court, exercising its inherent power under Section 482 of the Code of Criminal Procedure, carried into Section 528 of the Bharatiya Nagarik Suraksha Sanhita, may quash the proceedings where the parties have genuinely settled, following B.S. Joshi and the three-judge bench decision in Gian Singh, which classed matrimonial offences as private in character for this purpose. Proceedings may also be quashed without any settlement where the allegations, taken at face value, do not disclose the offence, and courts apply that branch strictly to omnibus allegations against the husband’s relatives that plead no specific act by the relative named.

Q: What do critics of Section 498A argue?

Critics argue that the offence’s structure invites overreach: it is cognizable and non-bailable, it extends to any relative of the husband, and a single complaint can therefore bring parents, married sisters and elderly relatives into a criminal process at the moment of a marriage’s collapse, with arrest, in the decades before the safeguard regime, following registration as a matter of course. They point to the low proportion of complaints ending in conviction as evidence that many complaints cannot have been genuine, and to the non-compoundable classification as forcing parties who have settled to keep litigating. The case has named institutional holders: the Save Indian Family Foundation, the Malimath Committee, the Law Commission’s 243rd Report and parliamentary committees, whose common centre has been compoundability with safeguards and discipline on arrest rather than repeal.

Q: What do defenders of Section 498A argue?

Defenders argue that the baseline is under-registration, not over-registration: the fifth National Family Health Survey found that 29.3 per cent of ever-married women aged eighteen to forty-nine had experienced spousal violence, and that only about fourteen per cent of women who experienced violence sought help from any source, against which annual registrations of about a lakh and a half are a small fraction of the conduct. They read the low conviction rate as the product of private evidence, witnesses turning hostile under family pressure, settlements and the criminal standard of proof, since an acquittal shows only that the offence was not proved, not that it was invented. They read the high charge-sheeting rate as complaints surviving police screening, and they oppose screening committees as filters that delay protection during the period of greatest risk.

Q: Is arrest automatic after a complaint under Section 498A?

No, and it has not been in law since 2014. Under the Arnesh Kumar directions, preserved when the procedural law moved to the Bharatiya Nagarik Suraksha Sanhita, whose Section 35 carries the arrest conditions and the notice of appearance, a police officer dealing with an offence punishable with up to seven years must record reasons against the statutory checklist before arresting, and must use a notice of appearance where arrest is not necessary for investigation, for preventing further offences, or for securing the accused’s presence. A magistrate must record satisfaction before authorising detention. In the districts covered by the guidelines the Supreme Court approved in Shivangi Bansal, an additional two-month cooling period with committee referral runs before any arrest. Practice varies by district, but automatic arrest is a description of the older system, not of the governing law.

Q: Who can be accused under Section 498A?

Only the husband of the woman and a relative of the husband. The class of relatives is wide, covering parents, siblings and relatives by marriage, and it carries no requirement that the relative live in the same household or even in the same city, which is why complaints commonly name several members of the husband’s family. The class is not unlimited: in U. Suvetha the Supreme Court held that a relative must be related by blood, marriage or adoption, so a woman alleged to be the husband’s girlfriend cannot be charged under the section. On the other side of the relationship, the protection is not defeated by the husband’s technical defences: in Reema Aggarwal the Court applied the provision where the man had gone through a ceremony of marriage and treated the woman as his wife although an earlier marriage subsisted.

Q: What must a Section 498A complaint allege to survive a quashing petition?

It must allege, against each accused, specific conduct answering the statutory definition: identified acts, on identifiable occasions, of a gravity likely to drive the woman to suicide or to cause grave injury or danger to life, limb or health, or harassment tied to an unlawful demand for property or valuable security. The quashing jurisprudence, from Geeta Mehrotra through Kahkashan Kausar to Dara Lakshmi Narayana, consistently strikes proceedings built on general and omnibus allegations, complaints that narrate a household’s unhappiness and attribute it to every relative named without saying what each person did. A complaint that pleads a dated course of conduct, with the demand’s words and occasions where the demand limb is invoked, and separates the husband’s acts from each relative’s acts, is the kind that survives; a list of names attached to a narrative of discord is the kind that does not.

Q: Can a Section 498A complaint be filed from the parental home?

Yes. In Rupali Devi v. State of Uttar Pradesh, decided in 2019, the Supreme Court held that a woman who has been subjected to cruelty and has left the matrimonial home to take shelter at her parental home may file the complaint at the place of shelter, because the effects and consequences of the cruelty continue there and the courts of that place have jurisdiction to entertain it. The complaint is therefore not confined to the courts of the matrimonial home’s territory. The practical significance is considerable, since the typical complainant has already left the matrimonial home when she complains, and a rule tying the complaint to that home’s courts would require her to litigate from the place she fled, against the household she accuses, at the moment of her least security.

Q: What is the cooling-off period in Section 498A cases after Shivangi Bansal?

It is a period of two months from the lodging of the first information report or complaint during which, under the Allahabad High Court guidelines framed in Mukesh Bansal and approved by the Supreme Court in Shivangi Bansal in July 2025, no arrest is to be made and no coercive police action taken against the named accused, while the matter stands referred to the district Family Welfare Committee. The committees are constituted under the District Legal Services Authority with at least three members, and the referral covers cruelty complaints, including those accompanied by offences punishable with less than ten years, with complaints disclosing injury of the gravest kind outside the scheme. The Supreme Court directed that the guidelines remain operational and be followed by the authorities, acting under its power to do complete justice in the case before it.

Q: Is Section 498A compoundable?

No. The offence is classified as non-compoundable, which means the parties cannot settle it by compounding in the trial court even where both sides wish to, and a prosecution can end before verdict only by withdrawal in the limited cases the code permits, by acquittal or discharge, or by quashing in the High Court on settlement. The classification has been reconsidered repeatedly and retained: the Malimath Committee recommended in 2003 that the offence be made bailable and compoundable, parliamentary committees recommended compoundability in 2005 and 2008, the Rajya Sabha’s Committee of Petitions took the same view in 2011, and the Law Commission’s 243rd Report of 2012 recommended compounding with the court’s permission after a cooling-off period of three months, while keeping the offence non-bailable. Parliament did not enact any of these, including in the recodification of 2023.

Q: Does an acquittal prove that a Section 498A complaint was false?

No. An acquittal records that the prosecution did not prove the offence beyond reasonable doubt on the evidence that survived to the end of the trial; it is not a finding that the complaint was invented, and the court’s order frequently says nothing about the complaint’s truth either way. The reasons trials fail in this field are mostly recorded nowhere in the statistics: witnesses turning hostile during years of pendency, complainants withdrawing under family or economic pressure, settlements that empty the prosecution of its witnesses, and the difficulty of proving a private course of conduct to the criminal standard. A complaint can also be false, and courts do record that conclusion in particular cases, including in quashing proceedings; but falsity is a case-specific finding, and no aggregate of acquittals can be converted into a rate of falsity without reading the files.

Q: Why do so many Section 498A cases end in compromise if the offence cannot be compounded?

Because the marriage’s litigation is a constellation, and the criminal case is settled as part of the whole. The same breakdown generates the cruelty prosecution, domestic violence proceedings, maintenance, divorce and custody cases, and when the parties negotiate an end they negotiate all of it together, with the criminal case’s closure as one term of the package. The formal route for that closure is a quashing petition in the High Court under the Gian Singh principles, and the Bureau’s tables also record thousands of cruelty cases each year, about fifteen and a half thousand in 2022, in a compounded or compromised column, which reflects courts recording the settlement-driven end of cases the classification says cannot be compounded. The figures are the statistical shadow of a settlement system the statute does not mention and practice cannot do without.

Q: What is the presumption under Section 113A of the Evidence Act?

Section 113A, inserted by the same amending Act of 1983 that created the cruelty offence and carried into Section 117 of the Bharatiya Sakshya Adhiniyam, provides that where a married woman commits suicide within seven years of her marriage and it is shown that her husband or a relative of her husband subjected her to cruelty, the court may presume, having regard to all the other circumstances of the case, that the suicide was abetted by the husband or the relative. The presumption is discretionary, and it does not arise from the suicide alone: the prosecution must first prove the cruelty to the criminal standard, and the court may still decline the inference on the whole circumstances. It is deliberately weaker than the dowry death presumption, where the court shall presume once the foundational facts are shown.

Q: Can anticipatory bail be granted in a Section 498A case?

Yes. Anticipatory bail, the direction that a person be released in the event of arrest, is available for the offence as for other non-bailable offences, on application to the sessions court or the High Court under Section 482 of the Bharatiya Nagarik Suraksha Sanhita. It was in an anticipatory bail application arising from a cruelty complaint that the Supreme Court decided Arnesh Kumar itself. Courts commonly protect the applicants while the application is heard where the complaint appears to sweep in relatives on general allegations, and the merits are assessed against the complaint’s specificity, the applicants’ roles and the likelihood of the investigation requiring custody. Anticipatory bail decides nothing about the complaint’s truth; it decides only that the accused’s liberty should not be taken before trial while the process runs, which, given the pendency figures, may be for years.

Q: Is there a time limit for filing a Section 498A complaint?

The general limitation period for an offence punishable with imprisonment of up to three years is three years, running from the offence, under the limitation chapter of the procedural code. In practice the limit rarely defeats a cruelty complaint, for two reasons. The courts treat cruelty as a continuing offence, so that a course of conduct keeps the offence alive and the period runs from the conduct’s course rather than from its first incident; and the code permits a court to take cognizance after the period where the delay is properly explained or the interests of justice require it, a discretion exercised with the offence’s domestic setting in mind. Limitation arguments therefore succeed on particular facts, where a complaint is confined to stale and discrete incidents, rather than as a general answer to a complaint of continuing conduct.

Q: How is cruelty proved in a Section 498A trial?

Ordinarily, on testimony. The central evidence is the complainant’s own account of the course of conduct, supported where available by her parents and siblings, by neighbours or others who witnessed incidents or their immediate aftermath, by medical records where injury was treated, and by letters, messages or financial records where the demand limb is charged. There is no requirement of independent corroboration as a matter of law, since the offence’s setting is private and the courts accept that the witnesses will usually be family members, whose evidence is scrutinised with care rather than discarded for interest. The prosecution must prove the conduct and its statutory gravity beyond reasonable doubt, and the characteristic failure of these trials is not the absence of a legal route to proof but the fragility of family testimony across years of pendency, as witnesses settle, resile or die before the evidence is recorded.