selling
How accurate are online house valuations? Our backtest of 600 sales
How accurate are online house valuations? Our backtest of 600 real sales: median error, how often estimates land within 10% and 20%, and why they miss.
- 5 min read
- Last reviewed 5 November 2026
- By Your Property Check team
Key answer
Part 1 of 8
What we tested
General information, not advice. How accurate are online house valuations? Most articles on this subject repeat opinions. We have numbers, because we test our own free house value check against sales that have already happened. This page reports those results, including the unflattering ones. We do not test or comment on any other company's tool, so nothing here says one provider is better or worse than another.
The test works like this. We take 600 homes that really sold, hide that sale, and ask the model for a value using only sales recorded before the start of that month. The index adjustments are held back two months, as in real use. We then compare the estimate with the price actually achieved. We use only Land Registry "category A" sales, which GOV.UK describes as a single residential property sold for value. That keeps repossessions, power-of-sale transfers and bulk buys out of the sample.
Part 2 of 8
The headline results
| Measure | July 2026 test | June 2026 test |
|---|---|---|
| Sales tested | 600 | 600 |
| Median error | 12.2% | 11.8% |
| Within 5% of the price achieved | 21% | 21.5% |
| Within 10% | 41.7% | 43.3% |
| Within 20% | 70.2% | 73.8% |
| Achieved price inside the range we showed | 75.8% | 81.2% |
| Median bias | -1.3% | -1.1% |
Source: data/backtests/backtest-2026-07.json and backtest-2026-06.json, our own tests. Median error means half the estimates were closer than this and half were further away. Median bias of -1.3% means our estimates sat very slightly below the achieved price in the middle of the pack.
Part 3 of 8
What a 12% error means in pounds
Take a home that truly sells for £300,000. An estimate that is 12.2% out is about £36,600 away, so the figure shown might be £263,400 or £336,600. That is the typical miss, not the worst one. In the July test, three in ten estimates were more than 20% away, which on the same home is more than £60,000. This is why we show a range and not a single "magic number", and why we never present an online figure as a formal valuation.
Part 4 of 8
Which homes are easier to estimate
| Property type | Sales tested (July) | Median error | Within 10% |
|---|---|---|---|
| Detached | 163 | 14.1% | 39.9% |
| Flat or maisonette | 81 | 15.7% | 32.1% |
| Semi-detached | 171 | 10.6% | 45% |
| Terraced | 185 | 11.5% | 44.3% |
Source: data/backtests/backtest-2026-07.json. In June the order differed (terraced homes were the closest at 9.7% median error, with 51.6% within 10%), which tells you the sample sizes per type are small and the gaps between types move from month to month. What stays true is that flats and detached homes were harder than semi-detached and terraced homes in both months. Detached homes differ most from their neighbours, and flats vary by floor, lease and building.
Part 5 of 8
Does a home's sale history help
We also ran a much larger test on repeat sales: 39,197 homes sold between January and June 2026 that had sold before. Using each home's own earlier sale as an extra clue, the median error depended heavily on how long the owner had held it.
| Years since the home's previous sale | Median error | Within 10% |
|---|---|---|
| Under 5 | 6.9% | 65.2% |
| 5 to 10 | 8.1% | 59.2% |
| 10 to 20 | 10.9% | 46.6% |
| Over 20 | 13.7% | 38.9% |
Source: data/backtests/repeat-sales-2026-10.json, the with-history slice, training on 2025 sales and testing on January to June 2026. The lesson is simple: the longer ago a home last sold, the less an automated figure knows about it, because the kitchen you fitted, the extension you built and the state of the roof sit between the old sale and today. Recently bought homes are easier to estimate.
Part 6 of 8
Why estimates miss
- They cannot see inside. Condition, layout, extensions and garden size do not appear in sales records.
- Sales data lags. GOV.UK says the time between a sale and its registration typically ranges from 2 weeks to 2 months, so the two most recent months of Price Paid Data are incomplete.
- Not every sale is in the data. GOV.UK lists exclusions, including right-to-buy sales at a discount, transfers subject to an existing mortgage, gifts and court-ordered transfers.
- Neighbours differ. A postcode can mix a tired flat with a refurbished one.
- Markets move. An index adjustment moves old sales on to today, but averages hide individual streets.
Part 7 of 8
How to cross-check any estimate
- Read the comparables. Open the real sold prices for your street and postcode on sold prices and compare like with like. Our article on where to find sold house prices lists the free sources, and the Price Paid Data explainer covers what the records contain.
- Ask two or more local agents. Citizens Advice says many agents do valuations for free, and it suggests asking two or more. A formal valuation by a surveyor usually carries a fee.
- Compare the spread. If three sources give figures within a few per cent of each other, the middle is a reasonable working number. If they are 20% apart, find out why.
- Use the working, not just the answer. An estimate that shows its comparables lets you spot a bad match, such as a bungalow used to value your terraced house.
For a wider look at the options, see our guide to how to find out how much your house is worth.
Part 8 of 8
When to use an estimate, and when not to
An estimate is fine for a first look: deciding whether to sell, checking your equity, or choosing which agent to call. It is not fine as the only basis for an asking price, a probate figure, a divorce settlement or a loan. Those need a formal valuation by a qualified surveyor.
This page is general information, not advice.
Our take
Judge an estimate by its working, not its number
A median error of 12.2% means a single figure is a starting point. The common mistake is to take one number from one tool as the price, when homes last sold long ago, flats and detached houses are the hardest to estimate, and the real answer sits in the nearby sales.
Our free house value check gives a range from real Land Registry sold prices in your postcode and shows the comparable sales and the working, with no email or phone needed. We publish our test results, including the misses, so you can see how far to trust it.
Our opinion, drawn from the facts on this page: not advice and not a forecast. We earn from OfferCheck reports and from introducing sellers to cash buyers, so weigh it with that in mind (see our disclosure). We are not authorised or regulated by the FCA.
Questions, answered
Frequently asked questions
How accurate are online house valuations?
Why is my online house valuation wrong?
Are online valuations better for some homes?
Is an online valuation the same as a formal valuation?
How can I check an online estimate?
Where this comes from
Sources
- Your Property Check: valuation backtests (July and June 2026, repeat sales October 2026)
- GOV.UK: About the Price Paid Data
- Citizens Advice: Selling a home
Last reviewed 5 November 2026 by the Your Property Check team. This page is general information, not legal, tax or financial advice. For your own situation, speak to a conveyancing solicitor or another qualified professional.