Chess Elo estimator: what your accuracy says
You can estimate a rating range from your accuracy, but not a rating. One game fits a wide range of players, and averaging more games narrows it only a little. These are Lichess ratings, not FIDE or Chess.com. Type your accuracy below to see where players with that score usually sit.
Type an accuracy to see which Lichess ratings it usually goes with. It runs in your browser.
No accuracy yet? Review a game first.
What one game can tell you
Not much on its own. Take an accuracy of 83 in Blitz. Of the Lichess players who scored it, the middle half were rated 1,450 to 2,000, and 8 in 10 were between 1,150 and 2,300. That range is about 1,150 points wide.
Accuracy and rating do move together. The link is weak for a single game: the correlation is 0.27 in Blitz, 0.31 in Rapid and 0.35 in Classical. Where 1 would mean a perfect match, these are low numbers.
Think of the result as a hint. It says your game looked like the games of players in that range. It does not say you are one of them.
At the edges it gets weaker. Under 60 or over 90, accuracy barely separates stronger players from weaker ones in our data. The calculator adds a warning there.
Average accuracy by rating
The other direction is cleaner. If you start from a rating, the typical accuracy is easy to read. In Blitz, the median goes from 67.9 in the 600–800 band to 85.6 in the 2600–2800 band.
The spread inside one band is large. The middle half of players sit 11 to 28 points apart. A weaker player can easily have a better game than a stronger one, and it happens often.
| Rating | Blitz | Rapid | Classical |
|---|---|---|---|
| 600–800 | 67.9 (51.6–78.9)n = 1,490 | 66.6 (49.8–77.7)n = 1,766 | too few |
| 800–1000 | 71.6 (58.7–80.9)n = 5,313 | 70.0 (55.8–79.9)n = 4,869 | too few |
| 1000–1200 | 74.0 (62.9–82.3)n = 10,284 | 73.0 (61.2–82.2)n = 8,927 | 71.2 (57.5–81.3)n = 311 |
| 1200–1400 | 75.7 (65.7–83.4)n = 14,408 | 75.6 (65.5–83.7)n = 11,956 | 74.8 (64.2–83.5)n = 575 |
| 1400–1600 | 77.4 (68.2–84.6)n = 18,523 | 77.5 (68.4–84.9)n = 14,388 | 77.9 (68.5–84.8)n = 992 |
| 1600–1800 | 78.5 (69.9–85.4)n = 18,289 | 79.3 (71.1–86.1)n = 13,997 | 81.3 (73.8–87.4)n = 1,124 |
| 1800–2000 | 79.8 (71.8–86.5)n = 15,947 | 81.0 (73.7–87.3)n = 10,737 | 82.9 (75.8–89.3)n = 837 |
| 2000–2200 | 81.3 (74.1–87.6)n = 11,154 | 83.1 (76.2–89.2)n = 5,771 | 86.3 (79.8–91.9)n = 309 |
| 2200–2400 | 83.4 (76.9–89.2)n = 7,096 | 85.8 (80.0–91.5)n = 2,053 | too few |
| 2400–2600 | 84.7 (78.6–90.4)n = 3,279 | 88.7 (83.2–94.4)n = 437 | too few |
| 2600–2800 | 85.6 (79.8–91.2)n = 993 | too few | too few |
Classical has the fewest games, only 4,330 player-games in all, and it has no band below 1000. Its numbers are rougher. At higher ratings, Rapid and Classical medians run a little higher than Blitz. Slower games leave more time to find good moves.
More games, better estimate
Averaging helps because it smooths out one bad day. We tested it on 924 active players with at least 10 games in our sample. For each number of games, we guessed their average rating from their average accuracy. The half-width below is how far off we were in 8 cases out of 10.
| Games averaged | 80% half-width | Blind guess |
|---|---|---|
| 1 | ±575 | ±592 |
| 3 | ±550 | ±592 |
| 5 | ±514 | ±592 |
| 10 | ±474 | ±592 |
A blind guess ignores accuracy and always names the middle of the group. With 10 games we beat it, but only from ±592 to ±474. After 5 games the gain slows. After 3 games it was ±550.
Be careful with these numbers. These players are active and often ask for analysis, so they are not typical. Your own spread may be wider. We will not claim a range tighter than ±400. The leak report averages your games for you and links back here.
How we did it
We used the Lichess open database for 2026-09, which is public domain (CC0). It holds many games, but only some carry the server analysis we need. We took the first 120,000 that did, which gives 240,000 player-games (each side of a game counts once).
- Both players had a rating.
- Each side made at least 10 moves.
- The server analysis covered at least 90% of the positions.
We turned each game's analysis into one accuracy per side with the same code the review uses. It follows the method on the Lichess accuracy page; see chess accuracy explained for the steps. We then grouped player-games by time control, sorted them by rating in 200-point bands and by accuracy in 5-point bins, and took percentiles. We left out bins with fewer than 100 player-games and left out Bullet, where accuracy tells you almost nothing about rating.
Our review runs a smaller engine in your browser, so we checked it against the server numbers on 18 games that had both (36 player-games). Our in-browser accuracy came out higher in 27 of 36 cases: about 2.3 points higher in the typical case, 3.2 on average. When you say the accuracy came from a review on this site, the calculator subtracts 2.3 points before it looks anything up. Accuracy from Lichess analysis is used as it is. That is a small sample, so treat the correction as rough.
Download the data
Free to reuse with a link to this page.
Limits
- Selection. Only games with a server analysis are included. Players who ask for one may care more than most.
- Not a random draw. We took the first 120,000 qualifying games in the 2026-09 file, which is sorted by time. They all come from the start of the month, so some times of day or time zones may be over-represented.
- Lichess only. The ratings are Lichess ratings. They are not Chess.com or FIDE ratings, and we do not convert between them.
- Few Classical games. Only 4,330 player-games, so those rows are rough.
- Engine differences. The table uses Lichess server analysis. Our in-browser engine searches differently, so the same game can score a little differently.
- Strong pairing effect. Lichess matches players of similar rating, so your opponent's strength is baked into the data.
FAQ
Can I work out my rating from my accuracy?
Only roughly. One game points to a range about 1,150 points wide, so it cannot tell a 1,150 player from a 2,300 player. Treat it as a hint, not a rating.
Does this work for Chess.com or FIDE ratings?
No. Every rating here is a Lichess rating. Lichess, Chess.com and FIDE ratings use different pools of players, so the numbers do not convert one to one. We have no checked conversion, so we do not offer one.
Why is the range so wide?
Accuracy depends on the game as much as on you. Opponent strength, the opening, how sharp the position is and the clock all move it. In our data, accuracy explains only a small part of why ratings differ (r = 0.27 in Blitz).
How many games do I need?
More is better, with limits. Averaging 10 games brought the 80% half-width down to about ±474, against ±592 for a blind guess. We would not claim anything tighter than ±400.
Which accuracy should I enter?
Use the accuracy from a game review here, or from Lichess, and pick which one in the calculator. Both follow the same published method; ours runs a few points higher, so we correct for it. A number from another site may sit on a different scale. See chess accuracy explained.
Explain My Chess is not affiliated with Lichess. The data comes from the Lichess open database (CC0). Lichess is a trademark of its owner.