Chess accuracy is a 0–100 score that measures how close your moves came to an engine's best play in one game. It is not a percentage of "correct" moves. Every move is scored by how much it lowered your winning chances (Win%), then those move scores are averaged into one game number. A move that keeps your Win% intact scores 100; a move that throws away 20 points of winning chances scores 40.
How chess accuracy is calculated
Our Game Review uses the method Lichess publishes on its accuracy page (fetched October 2026), run on Stockfish 17.1 in your browser. There are three steps.
1. Turn the evaluation into Win%. Stockfish gives a score in centipawns (100 = one pawn). We clamp it to ±1000 and convert it:
Win% = 50 + 50 × (2 / (1 + e^(−0.00368208 × centipawns)) − 1)
So 0.00 is 50%, +1.00 is 59.1%, +3.00 is 75.1% and +10.00 is 97.5%. Forced mates count as 100% or 0%. If you need a refresher on what those engine numbers mean, read how to read chess engine evaluations.
2. Score each move. Compare the mover's Win% before and after the move:
Move accuracy = 103.1668 × e^(−0.04354 × Win% drop) − 3.1669
The result is capped between 0 and 100. A move that improves your position cannot score above 100.
3. Combine the moves. We take two averages of a side's move accuracies and report their mean:
- a volatility-weighted mean, where each move is weighted by how much Win% swung around it (the standard deviation of Win% over a sliding window, clamped to 0.5–12), so moves in sharp positions count more than moves in dead ones;
- a harmonic mean, which punishes single very bad moves harder than a plain average. We floor each move at 1% inside it, so one 0% move cannot drag the whole game to 0.
Chess accuracy calculator: what each drop costs
This table is the move formula applied directly. Find how many Win% points a move lost, and read off its accuracy.
| Win% points lost | Move accuracy | Our label |
|---|---|---|
| 0 | 100.0 | Best or Good |
| 2 | 91.4 | Good |
| 5 | 79.8 | Inaccuracy |
| 10 | 63.6 | Mistake |
| 15 | 50.5 | Blunder |
| 20 | 40.0 | Blunder |
| 30 | 24.8 | Blunder |
The curve is steep at the start. Losing just 5 points already costs a fifth of the score, which is why a handful of small inaccuracies pull a game down further than most players expect. The labels and their thresholds are explained in blunder vs mistake vs inaccuracy.
Because the drop is measured in Win%, not pawns, the same material loss counts for less when the game is already decided. Dropping from +1.00 to 0.00 costs 9.1 points; dropping from +8.00 to +7.00 costs only 2.1, because both positions are already above 90%.
Accuracy in three famous games
We ran three famous 19th-century games, all long in the public domain, through Stockfish 17.1 (lite, single-threaded WebAssembly, the same build our site uses) at a fixed depth of 18 per position, then fed the evaluations into our accuracy module unchanged. "Best" means the move played was Stockfish's top choice at that depth.
| Game and side | Accuracy | Best | Good | Inaccuracies | Mistakes | Blunders |
|---|---|---|---|---|---|---|
| Opera Game 1858, White (Morphy) | 96.3 | 15 | 1 | 1 | 0 | 0 |
| Opera Game 1858, Black (Duke and Count) | 81.2 | 4 | 7 | 4 | 1 | 0 |
| Immortal Game 1851, White (Anderssen) | 77.7 | 16 | 0 | 5 | 1 | 1 |
| Immortal Game 1851, Black (Kieseritzky) | 67.4 | 8 | 6 | 4 | 2 | 2 |
| Evergreen Game 1852, White (Anderssen) | 88.1 | 15 | 5 | 3 | 0 | 1 |
| Evergreen Game 1852, Black (Dufresne) | 78.7 | 12 | 6 | 2 | 1 | 2 |
Method: each PGN was replayed and checked with chess.js; every position was searched to depth 18 from a fresh engine state, and checkmate positions were scored as won. Repeat it at another depth and individual labels near a threshold can flip.
Two lessons stand out. Morphy's Opera Game is near-perfect by modern standards: one inaccuracy in 17 moves. And the two Anderssen "masterpieces" both contain a move Stockfish calls a blunder. In the Immortal Game, 18.Bd6 drops White from about +2.4 to −1.4 at depth 18; at depth 24 Stockfish prefers 18.d4 at +2.80. In the Evergreen Game, the famous 17.Nf6+ falls from +3.68 to +0.77; at depth 22 Stockfish rates 17.Ng3 at +4.29 and 17.Nf6+ at +0.82. Both attacks won because the defenders went wrong in reply, which is exactly what accuracy captures and a result does not.
Start
Paste any of these games into Game Review to see the move-by-move graph. Our browser engine searches to its own depth, so expect numbers close to, not identical with, the table.
What is a good accuracy?
There is no official scale, but the formula tells you what the numbers mean in practice:
- 90 and above: almost every move kept your winning chances. Typical of short, clean wins or very strong play.
- 80–90: solid. A few inaccuracies, rarely a real mistake.
- 70–80: one or two mistakes or a blunder your opponent did not fully punish.
- Below 70: the game turned on errors, often on both sides, as in Kieseritzky's 67.4.
Judge accuracy against your opponent's score in the same game and your own average, not against a fixed number. A sharp, tactical game produces lower scores for both players than a quiet one.
Frequently asked questions
What is a good accuracy in chess?
Above 80 is a good game for most club players, and above 90 is excellent. Compare it with your opponent's score in the same game: in our sample, Morphy beat his opponents' 81.2 with 96.3.
How is chess accuracy calculated?
Each evaluation is converted to Win%, each move scores 103.1668 × e^(−0.04354 × drop) − 3.1669, and the game score is the mean of a volatility-weighted average and a harmonic average of those move scores. The full method is above and on the Lichess accuracy page.
Why is my accuracy different on Chess.com and Lichess?
They use different formulas. Chess.com uses its own CAPS2 algorithm, which it describes as designed to "replicate the feeling of being graded on a test in school," with most scores between roughly 50 and 95 (Chess.com help, as of October 2026). Lichess states plainly that "there is no universally accepted way to compute accuracy percentages." Engine version and search depth differ too. Our Game Review follows the Lichess method, so its scores are closer to Lichess than to Chess.com, but they still will not match either exactly.
Can accuracy tell my rating?
No. There is no official accuracy-to-Elo conversion from FIDE, Chess.com or Lichess, and any chart you see online is someone's estimate. Lichess notes that accuracy and rating are correlated "but not in a straightforward way": stronger players choose sharper lines where both sides err, and an opponent's early blunder can hand you a high score. One 95% game says more about the game than about your rating. Your rating comes from results: the Elo calculator shows how wins and losses move it. The coach turns your saved reviews into practice exercises.
Review your own games in Game Review, test the critical moments in the analysis board, and read how to analyze a chess game to turn the score into a plan.