6 in 10 land within 10% of it, and 86 in 100 within 20%. We tested our valuation model on 128,804 homes sold in England and Wales between May and July 2026, none of which it had seen before. This is how close it came, where it finds homes harder to value, and how it compares with the model it replaced.
Picture 100 homes. For 61 of them our estimate was within 10% of the price paid. For another 25, it was within 20%. For the last 14, we missed by more than that. On a £300,000 home, a typical miss of 7.7% is about £23,000 either side. The average miss is higher, 12.7%, because a few large misses pull it up. That is why we lead with the typical one.
We scored the model it replaced on exactly the same 128,804 sales. It landed 38 in 100 within 10% of the price paid. This one lands 61. Its range is 35% narrower too: about £126,000 from low to high on average, against £194,000, and it still holds 80.8% of sales.
Every estimate comes with a low and a high figure. We set that range so 4 in 5 sales should land inside it. In the test, 80.8% did. It holds at every price. In each band, from estimates under £147k to estimates over £597k, between 79.9% and 81.8% of sales landed inside the range. A range that caught every sale would be too wide to help anyone decide anything, so a figure close to the target is the aim, not a shortfall.
Flats are the hardest homes to value: the typical miss is 8.8%, against 7.2% for semi-detached homes. Cheaper homes are harder too. For estimates under £147k, the typical miss is 10.7%, against 6.7% where it does best. The same gap in pounds is a bigger share of a smaller price.
An automated valuation is not a survey: it cannot see a home's condition or a new kitchen. The figures on this page describe the model across tens of thousands of sales, not any one home, so treat each estimate as an evidence-based starting point and bring what you know about the home to it.
Recent completed sales across England and Wales, recorded by HM Land Registry, that the model had never seen. Scotland and Northern Ireland are not covered.
R-squared mostly reflects how widely house prices vary across the country, so it looks high for almost any model. The miss on each sale says more about how close an estimate is likely to be for one home, so we lead with that and show R-squared in the full figures.
Homes that are unlike their neighbours are harder to value, and cheaper homes carry bigger percentage errors for the same gap in pounds. The tables show where the model is strongest and weakest.
Yes. We publish the test results of each model that goes live, so when a new one replaces this one, the figures and the story on this page change with it.
Source: CheckMyStreet back-testing of CMS Property Valuation v6 against HM Land Registry sold prices, published under the Open Government Licence v3.0. Coverage is England and Wales.