You've found a house you like, so you type the postcode into the police crime map, and thirty pins land on the street. A couple of them look like they're sitting on the doorstep.

They aren't. Those pins have been moved on purpose, and the one outside the front door may be a crime that happened several streets away. The map is still worth looking at. It just isn't good at the thing most people use it for.

What the crime map is actually showing you

Street-level crime data on police.uk is anonymised before publication, because real coordinates would identify victims.

Every crime is snapped to the nearest point on a fixed national list of map points. Those points sit over the centre of a street, or over a public place like a park or an airport, or over a commercial premises like a shopping centre or a nightclub. The coordinates you see are the map point's, not the crime's.

That list was filtered. Every point has to cover a catchment containing at least eight postal addresses, or none at all. Points covering between one and seven addresses were removed, because a point covering three houses narrows it down too far. And if the nearest available point is more than 20km away, the coordinates are zeroed out and the crime isn't mapped anywhere.

A pin is not an address. It's the nearest permitted place to put a dot. police.uk says so itself: the street-level crimes are "only an approximation of where the actual crimes occurred, they are not the exact locations".

It also admits the underlying geocoding is uneven. "Inconsistent geocoding policies in police forces mean we cannot be confident that the location data provided is fully accurate or consistent," with accuracy estimates across forces ranging from 60% to 97%.

So a pin outside a particular house tells you nothing about that house. And a street with no pins on it isn't a street where nothing happened.

The categories hide more than they show

There are fifteen crime categories as at August 2026, and one of them is "Violence and sexual offences", as a single combined bucket. A shove outside a pub and a serious assault go into the same category and come out as identical dots.

Nothing in the published data records severity, or whether anyone was charged, or whether the people involved had any connection to the area at all.

"Anti-social behaviour" is its own category, and it's built from incidents reported to the police rather than from recorded offences. It measures who picks up the phone as much as it measures what happened.

Both of those pull in the same direction. A road with a busy pub, a takeaway and a night bus stop will always out-record the residential street behind it, even in the cases where the residential street is the one you'd think twice about walking down.

Recorded crime is partly a measure of policing

This is the part that changes how you read all of it.

The ONS publishes two measures and they don't agree. The Crime Survey for England and Wales asks households what's actually happened to them, including things they never reported. Police recorded crime counts what forces wrote down.

In the year ending March 2026, published on 23 July 2026, the survey estimated 9.6 million incidents of headline crime, with no statistically significant change on the year before. The police recorded around 5.2 million crimes over the same period excluding fraud and computer misuse, which was 1% down.

The gap isn't a mistake in either number. Only an estimated 36% of survey-measured crimes were reported to the police at all.

The ONS is blunt about what follows: "Police recorded crime is not considered a reliable indicator of overall crime trends because of changes in police activity and recording practices." Where it does earn its keep is lower-volume, higher-harm offences that do get reported, like homicide, offences involving weapons, and robbery.

Worth holding onto when the totals look like they're going the wrong way. Survey-measured violence with or without injury came in at 989,000 incidents in the year to March 2026, 33% below the 1.5 million estimated a decade earlier.

Why a "high crime" label always has a top quarter in it

Some of our own numbers here, because this is a mistake we had to design around ourselves.

We score every listing we hold against the area around it, and crime is one of those scores. The label we attach is a position, not a level. We compare the area's recorded crime rate per 1,000 residents against a London benchmark and call anything above the upper quartile high.

Which means roughly a quarter of everywhere gets called high, whatever the actual numbers look like. Across active listings in and around London in August 2026, about 30% of the ones we hold carry that label. That figure describes the shape of the scale. It doesn't describe danger.

The same arithmetic goes wrong more sharply at the other end. A rate per 1,000 residents falls apart where hardly anybody is a resident. Think of a retail park, a station approach, or a town centre with a handful of flats above the shops: the numerator counts everyone passing through and the denominator counts the few people who sleep there. Those areas can post rates that make them look like the worst places in the country. We hold them out as outliers instead, and they're about 5% of what we've got.

Two caveats that belong next to those figures. They're drawn from active asking-price listings in and around London and the commuter belt, not a national sample. And the crime figure attached to a listing is its LSOA's figure, an area of between 1,000 and 3,000 residents, not a reading taken at the front door.

None of this appears in the listing, and it isn't supposed to. Agents have no duty to tell you what the area is like, which is a wider pattern in how listings are written.

What to do about it

Zoom out before you zoom in. Set the neighbourhood against the borough, and the borough against the region. A count on its own is a number with nothing to lean on.
Read the categories, not the total. Vehicle crime, burglary and shoplifting tell you three different things about a street. Shoplifting stacked up on a parade of shops is not a reason to rule out the road behind it.
Ask what's inside the boundary. A station, a hospital, a stadium, a high street or a night bus route will lift an area's numbers regardless of the residential streets in it.
Look at a year, not a month. The public data covers a rolling 36 months, currently July 2023 to June 2026, and it runs roughly two months behind. One quiet month proves nothing.
Find the neighbourhood policing team. police.uk will tell you which team covers the address and what they've published as their local priorities. That's a far better description of what's going on than a pin count.
Then go and stand there. 8am on a weekday, 11pm on a Friday, and again when the schools come out. Nothing in any dataset captures what a street feels like.

The honest summary

Crime maps are good at density and bad at location. The pins are moved deliberately, the categories are too broad to tell a pub scuffle from something serious, and the totals track reporting and recording as much as they track crime. Use them to work out which questions to ask, and never to rank one street against the one next to it.

If a single number makes you write off a whole area, it's usually the scale talking rather than the place.

This is general information, not legal or safety advice. Figures are as at August 2026: police.uk street-level data was current to June 2026 at the time of writing, and the ONS figures come from the Crime in England and Wales bulletin for the year ending March 2026, published 23 July 2026. police.uk covers England, Wales and Northern Ireland. Scotland publishes its crime statistics separately and on a different basis.