Chess accuracy: what the percentage means
Chess accuracy is a score from 0 to 100 for how much winning chance your moves kept. A move that loses nothing scores about 100. Bigger drops score much lower. The game score averages them.
How we work it out
Our method follows the one Lichess publishes on its accuracy page. The window sizes and weight limits in step 3 are our own choices. There are three steps.
1. Engine score to win percentage
The engine rates each position in centipawns (hundredths of a pawn). We turn that into the chance of winning:
Win% = 50 + 50 × (2 / (1 + e^(−0.00368208 × centipawns)) − 1)
The score is capped at ±1000 centipawns, and a forced mate counts as 100% or 0%. Being a queen up and being two queens up look almost the same: you are winning either way.
2. Win percentage to move accuracy
For each move we compare your Win% before and after it:
accuracy = 103.1668 × e^(−0.04354 × drop) − 3.1669
The drop is how many Win% points you gave away, and the result is kept between 0 and 100. From an equal position, these drops give:
- Drop of 0 points: accuracy 100.0
- Drop of 5 points: accuracy 79.8
- Drop of 10 points: accuracy 63.6
- Drop of 20 points: accuracy 40.0
- Drop of 40 points: accuracy 14.9
Small slips cost little. Big ones are punished hard.
3. Move accuracy to game accuracy
We then combine the moves of one side in two ways and average the two results:
- Volatility-weighted mean. Moves in sharp, swinging stretches count more than moves in quiet ones.
- Harmonic mean. It pulls the score down when a few moves are very bad, so one blunder is not hidden by many fine moves.
The final number is rounded to one decimal.
An example: the Opera Game
In the Opera Game (Paris, 1858) White scores 98.2% and Black 82.8%. White has 5 great moves and no mistakes. Black has 4 inaccuracies, which is enough to pull the score down. You can open the review and see each move's label.
Why it will not match Chess.com
We could not find a full public description of how Chess.com computes accuracy, so we can only say this much: it may use a different engine, search depth, formula or averaging. The same game can get different numbers on two sites. Compare accuracy within one tool, not across tools.
What moves your accuracy
- Game length. In a short game, each move carries more weight in the average, so one slip shows up more.
- Lopsided positions. When you are already winning, Win% flattens out, so small inaccuracies cost little. Equal positions punish them more.
- Forced moves and book moves. They rarely lose winning chance, which lifts a game that is mostly opening theory.
- Your opponent. An opponent who plays well keeps positions sharp. One who blunders gives you easy moves.
- The engine's own noise. A search is not perfect, so a very small gap between two games is not meaningful.
What is a good accuracy?
We have no data on average accuracy by rating, and we will not invent a table. A better use of the number is trend: review your games and watch whether your own average climbs, then check the leak report to see which mistakes cost you most.
Accuracy questions
What is a good accuracy in chess?
There is no fixed answer. We have no data on average accuracy by rating, so we do not publish a table. Compare your own games over time rather than against a number.
Can I get 100% accuracy?
Yes, if every move matches the engine or loses almost no winning chance. Long, messy games rarely end at 100.
Why is my accuracy different on Chess.com?
We could not find a full public description of Chess.com’s method, so we cannot say exactly how it differs. Engine, search depth and averaging all change the result, so the numbers are not directly comparable.
Does accuracy equal rating?
No. Accuracy is a score for one game, measured against an engine. It says nothing about the strength of your opponent or the difficulty of the position.