What is a good accuracy in chess?

Accuracy is a percentage that summarises how close your moves were to the engine’s recommendations across a whole game. It is computed from how much each move changed your winning chances (win%), not from raw centipawns directly - a move that barely dents a winning position costs little accuracy, while one that swings the game costs a lot. The result is a single 0-100 number per side, per game.

Because it is win-percentage based, accuracy is forgiving of "only slightly worse" moves in decided positions and harsh on real turning points - which matches how games are actually won and lost.

There is no official scale, but rough real-world bands help. Beginners often land in the 60s-70s; improving club players commonly score high 70s to high 80s; experts and masters routinely reach the 90s in calmer games. Correspondence or slow games score higher than blitz, and sharp tactical battles score lower for everyone because there are more chances to go wrong.

The table below gives approximate bands, pairing accuracy with the average centipawn loss (ACPL) you would expect at each level. Treat them as observed rough guides, not fixed thresholds - they shift with time control and the sharpness of the position.

The same player scores very differently across time controls, so always compare like with like. Classical and correspondence games score highest because you have time to check for tactics; rapid sits in the middle; blitz and bullet score lowest because you are guessing more under the clock. As a rough rule, expect blitz to run several accuracy points below your rapid at the same rating, and bullet lower still.

The single most important caveat: one game’s accuracy is noisy. A single blunder can drop an otherwise clean game by ten points or more. Judge yourself on your average across many games within one time control, not on your best or worst single result.

ACPL is the older, more literal metric: the average number of centipawns (hundredths of a pawn) you lost per move compared with the engine’s best move. Lower is better - strong club players often average somewhere around 20-40 ACPL, though it varies a lot by time control and position type. Accuracy is essentially a friendlier, win-percentage-weighted repackaging of the same idea, scaled to 0-100 so it is easier to read.

The two usually move together: lower ACPL means higher accuracy. Accuracy is better for a quick read; ACPL is useful when you want the raw magnitude of your errors.

Not necessarily. Accuracy is easier to keep high in quiet, simple positions and harder in sharp ones, so a 78% in a wild tactical game can reflect stronger play than a 90% in a lifeless one. It also says nothing about whether you found the best plan - only how close your moves were to the engine. Use it as one signal among several, alongside where your evaluation actually swung.

Accuracy rises when you cut blunders and inaccuracies, so the fastest gains come from studying your own mistakes: find the openings and structures where your accuracy drops, drill the tactics you miss, and slow down at the critical moments the eval graph flags. Chasing a number game-by-game is a trap; changing the habits that cause the errors is what moves the average.

Audit Chess makes this concrete: it computes a per-game accuracy score and eval graph across your whole Lichess or Chess.com history, shows your accuracy trend, and turns your recurring blunders into puzzles so you can drill exactly what is dragging the number down. It is free, needs no account, and runs in your browser.