An inaccuracy, a mistake and a blunder are three sizes of the same thing: a move that lowers your winning chances compared with the engine's best move. In our Game Review, a move that costs 5–10 Win% points is an inaccuracy, 10–15 is a mistake, and 15 or more is a blunder. Anything smaller is Best (the engine's top move) or Good.
Our thresholds
Every evaluation is first turned into Win%, a 0–100 estimate of your winning chances; the formula is in chess accuracy. The label depends only on how many points the move dropped.
| Label | Win% points lost | From 0.00, roughly | From +3.00, roughly |
|---|---|---|---|
| Best | under 5, and the engine's top move | none | none |
| Good | under 5, not the top move | under 0.54 pawns | under 0.68 pawns |
| Inaccuracy | 5 to under 10 | 0.54 pawns | 0.68 pawns |
| Mistake | 10 to under 15 | 1.10 pawns | 1.31 pawns |
| Blunder | 15 or more | 1.68 pawns | 1.89 pawns |
The pawn columns are the evaluation drop that reaches each threshold, computed from our Win% formula. They change with the starting position because Win% flattens out as an advantage grows: you need to throw away more to lose the same winning chances when you are already ahead.
These thresholds are the ones in Lichess's open-source analysis code, which marks moves at drops of 0.1, 0.2 and 0.3 on its −1 to 1 winning-chance scale (as quoted from the code in this Lichess forum thread). That is 5, 10 and 15 points on a 0–100 scale. The Lichess accuracy page documents the Win% formula they rest on.
How Chess.com classifies moves
Chess.com uses a different scale. Its move classification help page (as of October 2026) measures expected points, where 1.00 is winning, 0.50 is even and 0.00 is losing, and bases them on both the engine evaluation and the player's rating.
| Chess.com label | Expected points lost |
|---|---|
| Best | 0.00 |
| Excellent | 0.00 to 0.02 |
| Good | 0.02 to 0.05 |
| Inaccuracy | 0.05 to 0.10 |
| Mistake | 0.10 to 0.20 |
| Blunder | 0.20 to 1.00 |
Chess.com's mistake and blunder bands are wider than ours: a blunder there starts at 0.20, which corresponds to 20 points on our 0–100 scale, against our 15. Its page also lists Great Move, Brilliant and Miss, which are based on what the move does rather than only on points lost. Because its expected points depend on rating and its curve is its own, the same move can get different labels on the two sites. We do not claim to match Chess.com. For the Brilliant label, see what is a brilliant move.
Labels in three famous games
We ran three public-domain classics through Stockfish 17.1 (lite, single-threaded WebAssembly) at a fixed depth of 18 per position and labelled every move with the thresholds above. Each PGN was checked with chess.js.
| Game and side | Moves | Inaccuracies | Mistakes | Blunders |
|---|---|---|---|---|
| Opera Game 1858, White | 17 | 1 | 0 | 0 |
| Opera Game 1858, Black | 16 | 4 | 1 | 0 |
| Immortal Game 1851, White | 23 | 5 | 1 | 1 |
| Immortal Game 1851, Black | 22 | 4 | 2 | 2 |
| Evergreen Game 1852, White | 24 | 3 | 0 | 1 |
| Evergreen Game 1852, Black | 23 | 2 | 1 | 2 |
Even the winners blundered. Anderssen's 18.Bd6 in the Immortal Game and 17.Nf6+ in the Evergreen Game are both blunders by the engine, and both games were won anyway. Labels near a threshold can change at a different depth, so read them as the size of the error, not a verdict on the game. The accuracy scores for the same games are in chess accuracy.
Worked examples: one of each
All evaluations below are Stockfish 17.1 at depth 18, from White's point of view.
Inaccuracy: 6...Nf6 in the Opera Game
rn1qkbnr/ppp2ppp/8/4p3/2B1P3/5Q2/PPP2PPP/RNB1K2R b KQkq - 1 6
Open the position before 6...Nf6 →
White's bishop already eyes f7, and Qb3 would add the queen behind it while also hitting b7. Before the move the evaluation is +1.17 (Black's Win% 39.4); after the natural 6...Nf6 it is +2.11 (31.5). That 7.9-point drop is an inaccuracy. At depth 24, Stockfish's top choices are 6...Qf6 (+1.07) and 6...Qd7 (+1.29), both of which cover f7 with the queen.
Mistake: 9...b5 in the Opera Game
rn2kb1r/pp2qppp/2p2n2/4p1B1/2B1P3/1QN5/PPP2PPP/R3K2R b KQkq - 1 9
Open the position before 9...b5 →
Black tries to kick the bishop while still undeveloped. The evaluation jumps from +1.43 to +2.96, a 12.0-point drop for Black: a mistake. Morphy answered 10.Nxb5, sacrificing a knight to rip open the lines.
Blunder: 20...Na6 in the Immortal Game
rnb1k1nr/p2p1ppp/3B4/1p1NPN1P/6P1/3P1Q2/P1P1K3/q5b1 b kq - 1 20
Open the position before 20...Na6 →
Black is two rooks and a bishop up but stands worse at +1.57. After 20...Na6, Stockfish announces mate in three: 21.Nxg7+ Kd8 22.Qf6+ Nxf6 23.Be7#. Black's Win% falls from 35.9 to 0, the biggest drop in the three games. Stockfish's best defence is 20...Ba6, still −1.56 for Black at depth 24 but far from lost.
Start
How to use the labels
- Fix blunders first. One blunder costs more accuracy than several inaccuracies and decides most club games.
- Look for the pattern behind each mistake. Most come from an undefended piece or an unseen check; the tactics lessons train exactly those.
- Treat inaccuracies as study material, not failures. They usually mark a better plan you did not know.
- Check the position, not just the label. Open the move in the analysis board, compare it with the engine's line, and see how to analyze a chess game for a full routine.
Frequently asked questions
How big is a blunder in centipawns?
There is no fixed centipawn size, because we measure blunders in lost Win%. From an equal position a blunder starts at a drop of about 168 centipawns (1.68 pawns); from +3.00 it takes about 189. Losing a whole piece is almost always a blunder; losing a pawn often is not.
Is a blunder always a lost game?
No. A blunder means your winning chances fell by 15 points or more, not that they reached zero. Anderssen's 17.Nf6+ dropped White from +3.68 to +0.77, still better, and he won when Black went wrong.
What is the difference between a mistake and a blunder?
Size. A mistake costs 10 to 15 Win% points, roughly one pawn from an equal position. A blunder costs 15 or more and usually hands over material or a forced attack. On Chess.com the bands are 0.10–0.20 and 0.20 or more expected points.
Run your own games through Game Review to see every label, then turn your saved mistakes into practice exercises with the coach. For how the engine evaluations behind these labels work, read how to read chess engine evaluations.