In short
Accuracy is a score from 0 to 100 for how closely your moves kept your winning chances compared with the engine's best moves. Every move gets its own accuracy, and the game's accuracy combines them, weighing your worst moves more. In our benchmark of real Chess.com games the median player under 1200 scored about 69 and players rated 2000 and up about 84, but one game says little about strength.
After a game review, the first number most players look at is their accuracy. It is a useful number, as long as you know what goes into it and what it cannot tell you.
1What accuracy measures
Before each of your moves the engine works out your chance of winning, from the evaluation. After your move it does the same again. If you played the best move, your chances stay where they were and that move scores close to 100. The more winning chance a move gives away, the lower its accuracy, and it falls quickly: a move that throws away a large share of your chances scores near zero.
Measuring winning chances rather than pawns is what makes accuracy fair. Dropping a pawn in a level position matters a great deal. Dropping a pawn when you are already a rook up barely changes your chances, so it barely costs accuracy.
2How the game's accuracy is worked out
Your game accuracy is not a plain average of your moves. It mixes two averages. The first gives more weight to moves played while the game was swinging, because those are the moments that decide it. The second is a harmonic mean, which is pulled down hard by a single very bad move. Together they make sure one blunder shows in the number, even in a game of otherwise good moves.
The method has the same shape as Lichess's. Our settings are stricter, fitted so that our numbers read like Chess.com's. Across 440 public Chess.com games, our accuracy came out on average 1.2 points above Chess.com's own, where Lichess's settings would have read 6.5 points above. The full comparison, game by game, is on the benchmarks page.
3What is a good accuracy?
It depends on your level and on the game. Here is the median accuracy our review gave the players in the benchmark's Chess.com games, by their Chess.com rating:
| Rating | Median accuracy | Players' games |
|---|---|---|
| under 1200 | 69 | 399 |
| 1200-1599 | 79 | 133 |
| 1600-1999 | 80 | 196 |
| 2000 and up | 84 | 152 |
Treat these as rough guides. Half of the players in each band scored above the median and half below, and the spread is wide. A quiet, one-sided game produces high accuracy for both players. A sharp game full of tactics produces lower accuracy even between strong players, because there are more ways to go wrong.
4Why your accuracy differs between sites
- Each site uses its own formula and settings. Lichess's own settings are more generous than Chess.com's.
- Sites search to different depths. A deeper search finds more problems in moves that looked fine, which usually lowers accuracy a little.
- Opening moves are treated differently. Moves still in the opening book are easy to play well, and whether they count changes the number.
- The engine and its version differ, and so its evaluations differ slightly in every position.
5How to use accuracy well
- Compare yourself with yourself. Your average over your last twenty games is a far better guide than any single game.
- Look at accuracy by phase. If your middlegame is fine and your endgames lose points, you know what to study.
- Read the grades behind the number. Accuracy is a summary of the same evaluations that produce the inaccuracies, mistakes and blunders.
- Do not chase a high number. Safe, passive play can score well and still lose games you could have won.