In short
The Elo rating system predicts each game's result from the rating difference, then moves ratings by how far the result beat or missed that prediction. Your expected score against an opponent is 1 / (1 + 10^((Rb − Ra) / 400)), and your new rating is R + K × (S − E), where S is your score and K sets how fast ratings move. Beating an equal opponent with K = 20 gains 10 points.
The Elo rating system is named after Arpad Elo, a physics professor and chess player who designed it. The United States Chess Federation adopted it in 1960 and FIDE in 1970. Almost every chess rating you will meet today is either Elo or a descendant of it, so understanding the original makes all of them easier to read.
The idea is simple. Every rating predicts how two players should score against each other. After each game, or each event, the ratings move by how far the actual result was from the prediction. Score better than expected and you gain; score worse and you lose.
1How does Elo work in one sentence?
Elo turns the rating difference into an expected score, compares it with the score you actually made, and moves your rating by a fixed multiple of the difference. Everything else is detail about how the expected score is calculated and how big the multiple, called the K-factor, should be.
2What is the expected score?
In chess a win scores 1, a draw ½ and a loss 0. Your expected score against an opponent is the average you would make against them over many games, according to the ratings. Two equally rated players are each expected to score 0.5. The higher-rated player is expected to score more, and the gap grows with the rating difference.
E = 1 / (1 + 10^((Rb − Ra) / 400))
Ra = your rating
Rb = your opponent's rating
E = your expected score, between 0 and 1The 400 is a scale constant. It means that every 400 points of difference multiplies the odds by ten: a player 400 points stronger is expected to score about ten times as much as their opponent. The table shows what the formula gives at common differences.
| Rating difference | Stronger player's expected score | Weaker player's expected score |
|---|---|---|
| 0 | 0.50 | 0.50 |
| 50 | 0.57 | 0.43 |
| 100 | 0.64 | 0.36 |
| 200 | 0.76 | 0.24 |
| 300 | 0.85 | 0.15 |
| 400 | 0.91 | 0.09 |
| 600 | 0.97 | 0.03 |
| 800 | 0.99 | 0.01 |
An expected score of 0.76 does not mean the stronger player wins 76% of games. It means they score 0.76 points per game on average, which could be made up of wins, draws and the occasional loss. Elo predicts points, not the share of each result.
3What do a win, a draw and a loss score?
The score S in the formula is the plain game result, however the game ended. Checkmate, resignation and a win on time all score 1 for the winner. Stalemate, agreement, repetition and the other draws all score ½ each. There is no extra credit for a quick or pretty win.
That second position is a useful reminder that the rating system only sees the result. Being a queen up earns nothing if the game ends in stalemate, and the ½ point it produces is rated exactly like a draw agreed on move 20.
4The Elo formula for a rating change
R' = R + K × (S − E)
R = your rating before
K = the K-factor
S = your score (1, ½ or 0, summed over the games)
E = your expected score, summed over the same games
R' = your new ratingIf you score exactly what was expected, S − E is zero and your rating does not move. Each point you score above expectation is worth K rating points, and each point below costs K. In pure Elo with the same K for both players, the winner gains exactly what the loser loses.
5Worked examples with real numbers
These examples were calculated with the same code as the site's Elo calculator, with a K-factor of 20 unless stated otherwise. The changes are shown to two decimals before rounding.
| You (rated 1500) | Opponent | Expected score | Result | Change |
|---|---|---|---|---|
| Win against an equal player | 1500 | 0.50 | 1 | +10.00 |
| Win against a stronger player | 1700 | 0.24 | 1 | +15.19 |
| Draw against a stronger player | 1700 | 0.24 | ½ | +5.19 |
| Loss against a stronger player | 1700 | 0.24 | 0 | −4.81 |
| Win against a stronger player, K = 40 | 1700 | 0.24 | 1 | +30.39 |
Notice how lopsided the 1700 game is. Losing to a stronger player costs little because a loss was the likely outcome, while a win gains a lot because it was not. A draw against a stronger opponent is a rating gain, since you scored ½ when the formula expected about a quarter of a point.
Now a five-round tournament for a player rated 1600, again with K = 20. Each game's expected score is worked out separately, then the totals are compared.
| Round | Opponent | Result | Expected score | Change |
|---|---|---|---|---|
| 1 | 1500 | Win | 0.64 | +7.20 |
| 2 | 1650 | Draw | 0.43 | +1.43 |
| 3 | 1700 | Loss | 0.36 | −7.20 |
| 4 | 1550 | Win | 0.57 | +8.57 |
| 5 | 1800 | Draw | 0.24 | +5.19 |
| Total | Average 1640 | 3 points | 2.24 | +15.19 |
The player scored 3 points against an expected 2.24, so S − E is 0.76 and the change is 20 × 0.76 = 15.19. The performance rating for the event, the rating at which 3 out of 5 against these opponents would have been exactly expected, comes out at about 1717.
6What is the K-factor in chess?
K decides how far one result moves a rating. A high K makes ratings react quickly, which suits new players whose first number may be far from their real strength. A low K makes ratings stable, which suits established players whose strength changes slowly. Different systems choose K differently.
| FIDE situation | K-factor |
|---|---|
| New to the rating list, until 30 games are completed | 40 |
| Junior, until the end of the year of their 18th birthday, while rated under 2300 | 40 |
| Rated under 2400 | 20 |
| Once a published rating has reached 2400, even if it later drops | 10 |
FIDE adds a cap: if K multiplied by the number of games in a rating period would exceed 700, K is reduced so that it does not. A player with K = 40 who plays 20 games in a month therefore has K = 35 for that month. US Chess takes a different approach and calculates K from how many games your rating is based on, so it shrinks smoothly as your rating becomes more established.
7How FIDE ratings differ from textbook Elo
FIDE uses the Elo method but with its own details. Expected scores come from a published table rather than the formula, rounded to whole percentages. For players rated below 2650 in standard chess, a rating difference larger than 400 points counts as 400. The change for a rating period is rounded to a whole number.
In practice the results are very close to the formula. Run the five-round tournament above through FIDE's rules with K = 20 and the expected scores read from the table total 2.24, the change is 15.2 and the rounded result is +15, giving a new rating of 1615.
- Lowest published rating: 1400. A player whose rating falls below that is shown as unrated on the next list.
- First rating: based on at least five games against rated opponents, worked out from their average and the score, and capped at 2200.
- Rapid and blitz: from 1 December 2024, a game between players 600 or more points apart is not rated if either is rated above 2600.
8What is a provisional rating?
A rating based on few games is unreliable, so most systems treat it differently. US Chess considers a rating established once it rests on more than 25 games, uses a special formula for players with eight games or fewer, and has an absolute floor of 100. Rating sites often show a question mark, or the word provisional, next to a rating they are still unsure about.
9Elo vs Glicko: what is the difference?
Glicko, designed by the statistician Mark Glickman, keeps a second number alongside the rating: the rating deviation, which measures how unsure the system is. A new or inactive player has a large deviation, so their rating moves a lot. A regular player has a small one, so it moves less. Glicko-2 adds a third number, the volatility, which tracks how erratic a player's results are.
| Elo | Glicko-2 | |
|---|---|---|
| Numbers per player | Rating | Rating, deviation and volatility |
| How fast a rating moves | Fixed by K | Depends on how certain the system is |
| Inactivity | No effect | Deviation grows, so the next results count more |
| Used by | FIDE, and US Chess in a modified form | Lichess, and this site's puzzle ratings |
Both systems share the same core: an expected score from the rating difference and an update from the gap between expected and actual. The puzzle trainer on this site rates every attempt as a game between you and the puzzle, the way Lichess rates its puzzles, starting from 1500 with a large deviation and showing a question mark until the deviation has fallen far enough.
10Common misunderstandings about Elo
- A rating is not a measure of absolute skill. It only compares players in the same pool.
- Beating a weaker player is not worthless, it is simply expected. You gain a few points for doing what the rating predicted.
- Rating points are not lost for good. They move between players; in pure Elo with equal K, the total in a pool stays the same.
- A higher rating does not guarantee a win. It predicts a higher average score over many games.
- Two systems with the same name can give different numbers. FIDE Elo and a site's Elo are different pools.