ChessGrader / Guides / How to analyze your games

How to analyze your chess games

A five-step method that takes about twenty minutes per game — think first, engine second, and turn every mistake into a named pattern you can actually fix.

By the ChessGrader team · Updated July 10, 2026

Analyzing your own games is the oldest improvement advice in chess, and it's survived because it works. Botvinnik built the Soviet school around it; every modern coach repeats it. The reason is simple: your games are a perfect record of your specific mistakes. A tactics book teaches you patterns someone else missed. Your last blunder teaches you the pattern you personally miss, in a position you personally reach. No other study material is that targeted.

The catch is that most people "analyze" by clicking through an engine report, nodding at the blunders, and closing the tab. That's reading, not analysis. Here's a method that makes the twenty minutes actually stick.

Step 1: Annotate from memory, before the engine

Replay the game without any engine and write down what you were thinking at the moments that felt important. Where did you think you stood better or worse? Which move took you the longest, and what were you weighing? Where did you feel the game slip?

This step is the whole trick. The engine will tell you what the truth was; your notes tell you what you believed. Improvement lives in the gap between those two. Skip this step and the engine's verdicts have nothing to correct — you'll agree with everything and retain nothing.

Step 2: Find the 2–3 critical moments

You don't need to understand all forty moves. Most games are decided at two or three points: where the evaluation swung, where you had a choice between two plans, where the character of the position changed. Mark those. If you're using a tool with an eval graph, the big cliffs on the graph are your shortlist — but check it against your own list from step 1. A moment that felt critical to you but wasn't (or the reverse) is itself a finding.

Step 3: Check with an engine — and demand the why

Now turn the engine on, but only at your marked moments. For each one, don't stop at "the engine prefers Nf5." Ask why. What does Nf5 threaten? What did your move allow? Play the engine's line forward a few moves until the point is concrete — a fork appears, a file opens, a pawn falls. If you can't state the reason in one sentence, you haven't finished the moment. An engine verdict you can't explain is a verdict you'll repeat.

Step 4: Name the pattern

Classify each mistake in plain words. Was it a missed tactic (you didn't see the knight fork)? Plan drift (you started an attack, then grabbed a pawn on the other wing)? A calculation error (you saw the line but misjudged the end of it)? Time trouble (the move was fine until move 35, when you had ten seconds)? An opening gap (you were out of book on move 6 and it cost you)?

Keep a running list across games. After ten games the list stops being trivia and becomes a diagnosis: "I lose to tactics on the long diagonal" is fixable in a way "I blundered" never is.

Step 5: Drill the mistakes

Seeing the right move once isn't learning it. Set the position up again later and make yourself find the correct move cold — that's the difference between recognizing an answer and producing one. ChessGrader builds this in: every review includes mistake drills that replay your errors until you'd find the right move at the board. If you use another tool, do it manually: save the FEN, revisit in a week.

The method at a glance

What to skip

• Memorizing engine lines. The engine's 6-move refutation is not knowledge you'll ever use; you'll never see that exact position again. Extract the idea, discard the moves.
• Obsessing over accuracy. Accuracy is a summary statistic, not a study plan. It varies with time control and how one-sided the game was, and grinding it from 84 to 86 tells you nothing about what to fix. Here's what accuracy is actually good for.
• Analyzing every move. Forty shallow verdicts teach less than three understood ones.

A worked example

A 1200-rated rapid game. In step 1, our player notes: "Around move 18 I felt my attack was winning. Move 24, I couldn't see a plan so I traded queens." Step 2, the eval graph shows two cliffs: move 19 and move 26. Interesting — move 24, the trade that felt significant, was actually fine.

Step 3, move 19: the engine wanted Rf3, lifting the rook into the attack, and calls the played move (a3, a "useful waiting move") an inaccuracy. Playing the engine line forward, the point becomes concrete: Rf3, Rh3, and mate threats arrive two moves before Black's counterplay. The attack wasn't winning yet — it was winning only if every move counted. Move 26: a knight fork on e5 was available and missed; instead a pawn trade let the game drift to a draw. Step 4, the log gets two entries: "attacked slowly while feeling 'winning'" (plan drift) and "missed knight fork, central square" (tactic). Step 5, both positions go into the drill queue. Total time: nineteen minutes, two named weaknesses. That's a good analysis.

Which tools to use

The method is tool-agnostic. The honest one-liners:

• Lichess — free unlimited server analysis, strong and no-nonsense. Marks inaccuracies, mistakes, and blunders, but no Brilliant/Great/Miss-style labels and no eval-drill loop.
• Chess.com — the most polished review with coach commentary, but one free review per day, with the rest behind a paid membership.
• ChessGrader — free, unlimited, no signup; full label set, published methodology, and built-in mistake drills. Runs in your browser, imports from chess.com and Lichess or any PGN.

We've compared the whole field in the best free game review tools if you want the long version.