Chess Accuracy Calculator
Paste a PGN and get an accuracy percentage for both players, calculated the way modern chess sites calculate it: from how much win probability each move gave away.
What chess accuracy actually measures
Every position has a win probability attached to it, derived from the engine evaluation. A move that drops your win probability from 62 percent to 60 percent has cost you two points of win chance. A move that drops it from 62 percent to 15 percent has cost you forty seven.
Accuracy converts that loss into a 0 to 100 score with a curve, so small losses cost almost nothing and large ones cost a lot. Two things follow, and both surprise people. First, a game full of small imprecision can still score very high. Second, one genuine blunder can drag a whole game down, because the curve is steep at the wrong end.
The method, in full
For each move the calculator does the following:
- Search the position before the move and record the evaluation.
- Search the position after the move and record the evaluation, from the mover's point of view.
- Convert both to win probability and take the difference. That is the win chance loss for the move.
- Map that loss onto a 0 to 100 accuracy score with a decay curve.
The per move scores are then combined into one figure per player, weighted so that moves in volatile positions count for more than moves in dead ones. This is the same computation the Chess DNA app performs, running the same code, which is why the number you get here is the number you would get in the app at the same depth.
Why depth changes the score
Accuracy is derived from evaluations, and evaluations move with search depth. A shallow search can miss a resource three moves later and mark a good move as a mistake, which pushes the score down. A deeper search is steadier.
So an accuracy figure is only comparable to another figure computed at the same depth and with the same engine. Comparing a depth 12 browser score with a number from a different site is not meaningful. Chess DNA runs every imported game at depth 18 for exactly this reason: one consistent setting, so the trend across your games is real rather than an artefact of the settings.
What counts as a good accuracy?
Honest answer: it depends heavily on the position type, the time control and the depth used, so no single threshold is meaningful across games. Sharp tactical games score lower than quiet ones for everybody, including grandmasters, because there is more to give away.
A far more useful reading than the number itself is the moves that cost the most table underneath it. Accuracy tells you that something went wrong. That table tells you what, and it is the part worth studying.
Frequently Asked Questions
How is chess accuracy calculated?
Each move is converted into a win probability loss by comparing the evaluation before and after it, and that loss is mapped onto a 0 to 100 score through a decay curve. Those per move scores are then combined into a single figure per player, weighted towards positions where the evaluation was volatile.
Is accuracy the same as the percentage of best moves?
No, and this is the most common misunderstanding. A player can find no engine first choices at all and still score highly if none of their moves gave away meaningful winning chances. Conversely, matching the engine for thirty moves and then hanging a queen produces a low score.
Why do two sites give different accuracy numbers for the same game?
Different engines, different search depths and different curves. Accuracy is a model, not a physical measurement, so a figure is only comparable to another figure produced the same way.
What is a good accuracy score in chess?
It varies too much by position type, time control and search depth for a single threshold to be honest. Within one setting a higher number is better, and comparing your own games to each other at the same depth is much more informative than comparing to a fixed target.
Is the accuracy calculator free?
Yes, with no account and no limit. The engine runs in your browser, so nothing is uploaded and there is nothing to pay for.
