Not what the data shows. The Crime Survey put headline crime at 4.3 million incidents in the year ending March 2026 — down on the year and less than a quarter of its 1995 peak — though adding fraud roughly doubles the total and fraud is not falling.
Figures in the text were correct on 4 September 2026; the charts below are drawn live from the same series and will move as new periods are published.
The best measure of long-run crime trends is the Crime Survey for England and Wales, because it counts what people say happened to them whether or not it was reported to the police. It estimated 4,333,000 incidents of headline crime (excluding fraud and computer misuse) in the year ending March 2026 — down from 4,595,000 the year before, and less than a quarter of the 19,786,000 estimated at the survey's 1995 peak.
Two honest complications. First, fraud: adding fraud and computer misuse takes the year-ending-March-2026 estimate to 9,615,000, slightly up on 9,445,000 the year before — fraud now accounts for around half of all crime against individuals, and it is not falling the way traditional crime has. The survey only began measuring fraud in 2016-17, so no 1995 comparison exists. Second, police-recorded crime (5,235,213 offences excluding fraud in the year ending March 2026, down from 5,307,433) measures police activity and recording practice as much as crime, so it cannot carry the long-run comparison on its own.
Individual offence types can rise while the total falls — theft and violence have their own pages with their own trends. "Crime" as a total, on the survey that exists to measure it, is close to the lowest level ever recorded.
Crime Survey estimate of all crime (excluding fraud)
Published valueProvisional or revisedLatestShaded bands: governments (see labels)Hover or use arrow keys to read a value
Source: ONS, Crime in England and Wales: Appendix tables.
How claims are checked
A verdict here says what the named official series show, on the periods named in the text — nothing more. Where two reasonable measures disagree, the verdict is “Depends which measure you use” and both are shown. Every series links to its full page with the publisher, the exact source cell, revisions and caveats. Free to reuse with attribution to Civometry and the original publisher.