Elo ratings explained: how chess and esports rank players

By StratPollen Studio · Updated 5 November 2026 · 5 min read

Win percentage tells you how often you win. It doesn't tell you who you beat. Going 7 out of 10 against club beginners is not the same as 7 out of 10 against strong players, but both records say 70%.

The Elo rating system fixes that by giving every player a number that rises and falls depending on the strength of the opponent. It was designed for chess by the physicist Arpad Elo, and ideas from it now run the rankings in online chess, many esports and plenty of other games.

A rating is a prediction

An Elo rating is not a score you collect. It is a prediction of how you will do against other rated players. The bigger the gap between two ratings, the more confidently the system expects the stronger player to win.

That prediction is called the expected score. A win counts as 1, a draw as ½ and a loss as 0, so an expected score of 0.76 means "this player should score about 76% against that opponent over many games".

The expected score formula

Expected score for A = 1 ÷ (1 + 10^{(Rating B − Rating A) ÷ 400})

You don't need to work through the exponent by hand. The table shows what it gives for common rating gaps:

The 400 in the formula sets the scale: every 400 points of difference multiplies the stronger player's odds by ten.

How ratings change after a game

After each game, your rating moves by the difference between what you actually scored and what the system expected, multiplied by a number called K:

New rating = Old rating + K × (Actual score − Expected score)

Worked example

Ana is rated 1600 and plays Ben, rated 1800. The gap is 200 points, so Ana's expected score is 0.24 and Ben's is 0.76. Both have K = 20.

• Ana wins: 20 × (1 − 0.24) = +15.2. Ana goes to about 1615 and Ben drops to about 1785.
• They draw: 20 × (0.5 − 0.24) = +5.2. Ana gains about 5 points and Ben loses about 5.
• Ben wins: 20 × (0 − 0.24) = −4.8. Ana loses only about 5 points, because losing to a stronger player was expected.

Notice the pattern. An upset moves ratings a lot, and an expected result moves them a little. When both players use the same K, whatever one gains the other loses.

What K does

K controls how fast a rating reacts. A high K lets a rating catch up quickly with a player who is improving. A low K keeps an established rating steady so one bad tournament doesn't send it tumbling.

Under FIDE's current rating regulations for standard chess, K is:

• 40 for a player new to the rating list, until they have completed events with at least 30 games;
• 40 for young players until the end of the year of their 18th birthday, as long as their rating stays under 2300;
• 20 while a player's rating stays under 2400;
• 10 once a player's published rating has reached 2400, even if it later drops below.

FIDE also changed its rules in March 2024, including raising the lowest published rating (the "rating floor") to 1400. If you look at older ratings, keep in mind they were calculated under slightly different rules.

Elo versus win percentage

The two numbers answer different questions:

• Win percentage describes your results: how often you won, drew and lost.
• Elo estimates your strength: how well you'd do against anyone, given who you actually played.

You can turn a score into a rough "performance rating" by reversing the formula. Take the average rating of your opponents and add the gap that matches your percentage:

Performance ≈ Average opponent rating + 400 × log₁₀(p ÷ (1 − p))

Say you score 7 out of 10 (p = 0.70) against opponents averaging 1700. That gives 400 × log₁₀(2.33) ≈ 147, so your performance was about 1847. FIDE uses its own lookup table for this, which gives very similar numbers. A perfect 100% or 0% has no finite answer, which is why tables cap the gap at around 800 points.

Online chess and esports

Most big online platforms use refinements of Elo rather than the original. Lichess uses Glicko-2 and Chess.com uses Glicko. Both add a "rating deviation", which is a measure of how sure the system is about your number. A new or returning player has a wide deviation, so their rating moves faster until the system has seen enough games.

Competitive video games usually keep a hidden matchmaking rating (often called MMR) built on similar ideas, then show players a rank or tier on top. The exact formulas are rarely published, but the principle is the same: beat stronger players and you climb faster.

One side effect matters for anyone tracking their record. Because matchmaking pairs you with opponents of similar rating, most players' win rates in ranked games settle close to 50%. That doesn't mean they've stopped improving. It means the system is doing its job. Our guide to what makes a good win rate goes into this.

Quick takeaways

• A 200-point gap means the stronger player is expected to score about 76%.
• Upsets move ratings a lot; expected results move them a little.
• K sets the speed: 40 for new and young players in FIDE chess, 20 for most, 10 at the top.
• Use win percentage to describe results and Elo to compare strength across different opponents.