Petrov Defense: What Each Rating Plays After 2...Nf6, and How to Punish the Stafford

The Petrov Defense is the opening Black chooses when the goal for the game is a safe one. After 1.e4 e5 2.Nf3 Nf6, Black does not defend the e5 pawn. Black counterattacks the e4 pawn, and if White takes on e5, Black simply takes back. It has a famous reputation: world champions Smyslov, Petrosian, Karpov and Kramnik have all used it, as has Caruana, and Nepomniachtchi played it against Carlsen in Game 4 of the 2021 world championship. The pages we opened tell that story well. They say little about what you actually meet when you play it, or when you face it, at a club rating. We ran the maia3-5m checkpoint, see how Maia chess actually works, on two positions at seven rating bands from 1000 to 2200 and graded every legal move with Stockfish 18 at depth 20. Percentages on this page are predictions from a human move model trained on online games. They are not real game statistics and they are not win rates.

Every evaluation is from White's side, so +0.50 means White is better by half a pawn. The two dataset positions are the one after 2...Nf6 with White to move, and the one after 3.Nxe5 Nc6 (the Stafford Gambit) with White to move. The line checks are separate local runs of ten seconds per line, which we call Stockfish on our machine, so a number can differ by several hundredths from other methods.

What the Petrov is, and why it is called Russian

The defining position is 1.e4 e5 2.Nf3 Nf6, ECO C42 and C43. White has three main third moves: 3.Nxe5, 3.d4 and 3.Nc3, and 3.Nc3 can transpose to the Four Knights Game. Wikipedia says the opening was popularised by Alexander Petrov, a Russian player of the mid 19th century, and also credits early analysis by Carl Jaenisch, which is why it is also called the Russian Game. Chess.com adds that it appeared in Pedro Damiano's work around 1500.

The reputation for draws is real at the top, and the sources say so plainly. Wikipedia writes that the Petrov has a drawish reputation, and a Lichess blog we saw only as a search snippet calls it equal, drawish, and not too exciting. Chess.com's verdict is more balanced: unless White is happy with a draw before the game begins, it is a strong opening for Black. In the 2018 world championship, Game 6 between Carlsen and Caruana was a Petrov, an 80 move draw. According to Wikipedia, at the 2004 Classical World Championship Kramnik beat Leko with White in both games where Leko played it, so the reputation has limits even at the top.

The first decision: what White plays after 2...Nf6

This is the position at the heart of the first table. Stockfish ranks 3.Nxe5 first at +0.50, then 3.d4 at +0.34 and 3.Nc3 at +0.19. 3.Be2 is +0.00, 3.d3 is -0.08, and 3.Bc4 is -0.58. Here is what the model predicts people play.

At 1000 and 1200 the likeliest move is the quiet 3.Nc3, at 32.2 and 32.3 percent, with 3.Nxe5 almost level at 30.0 and 31.8. From 1400 the engine's choice 3.Nxe5 leads, and the lead grows: 34.2 percent at 1800, 39.8 at 2000, and 47.6 at 2200. Meanwhile 3.Nc3 slides from 32.3 to 25.3, and 3.d4, which Wikipedia says was Steinitz's preferred move in 1889, rises from 5.1 percent at 1200 to 12.6 at 2200.

The line that deserves attention is 3.Bc4. It is the worst of the five by a wide margin, 1.08 pawns behind 3.Nxe5, and it is predicted about a fifth of the time at every band from 1000 to 1800, 19.6 to 21.0 percent. It then falls to 15.2 at 2000 and 9.3 at 2200. The reason is plain: it looks like an Italian Game move, but it leaves e4 undefended, and Black simply takes it with 3...Nxe4. Our check gives 3.Bc4 Nxe4 as -0.56 for White. Black is better, though not by a decisive margin, so this is a slip in initiative more than a blunder. The move belongs in the Italian Game, where Black has played 2...Nc6, and the Italian Game page covers that setup.

The practical reading is by band. Below 1400, expect a quiet 3.Nc3, which our check says leaves Black fine: after 3.Nc3 the best Black reply is 3...Nc6, the Four Knights, at +0.08 for White. 3...Bb4 is +0.35 and 3...d6 is +0.53, so 3...Nc6 is the clean answer. From 1400 up, expect the main 3.Nxe5, and from 2000, 3.d4 becomes a real option at 10.2 and 12.6 percent.

The main line: 3.Nxe5 d6 4.Nf3 Nxe4

After 3.Nxe5 Black plays 3...d6, not 3...Nxe4. White retreats 4.Nf3, and only then does Black capture on e4. This is the position every guide describes, and it is calm. Our check gives White three reasonable moves in a narrow band.

• 5.d4. +0.39, best at depth 26. The Classical Attack with 5...d5 6.Bd3 is the usual continuation, and chess.com lists 6...Be7, 6...Bd6 or 6...Nc6 as the most common replies.
• 5.Nc3. +0.32. After 5...Nxc3 6.dxc3 this is the Nimzowitsch Attack, where White accepts doubled pawns for fast development and 6...Be7 is the best and most common reply, according to chess.com.
• 5.Qe2. +0.22. The Lasker Variation, which Wikipedia says forces simplification and a likely draw after 5...Qe7 6.d3 Nf6.

The whole range from +0.22 to +0.39 is within about 0.17 of a pawn. That is the drawish reputation in numbers: no fifth move for White gets much. As of October 2026, chess.com's figure for 5.Qe2 is a draw in about four of every five master level games, and that is the only game statistic we found for the opening.

There is one trap on the main line worth knowing. After 5.Nc3, Black must avoid 5...Bf5 because 6.Qe2 pins and wins material. Chess.com reports that Anand lost to Alonso Zapata this way at Biel in 1988 and resigned after a handful of moves. Sources differ on the exact move count, so we do not quote one.

If you want other ways to meet 1.e4 e5 from the White side, the Scotch Game page and the Ruy Lopez page use the same method. They cover the 2...Nc6 replies to 2.Nf3, which is a different game from the Petrov.

The trap after 3.Nxe5: the greedy 3...Nxe4

The trap that catches the most players comes one move too early. After 3.Nxe5, a Black player who has learned the pattern "take on e4" plays 3...Nxe4? at once. White plays 4.Qe2, and Black is worse. Wikipedia calls 3...Nxe4 generally considered an error, and chess.com says it loses at least a pawn. We checked rather than copy that.

• 4...Qe7. This is Stockfish's best defence, and it leaves White only +0.67. Black is worse, but this is not simply a lost pawn. The follow up 5.Qxe4 d6 6.d4 gives +0.62 at depth 26.
• 4...d5. +1.41 for White. 4...d6 is +1.48.
• 4...Nf6?? 5.Nc6+ wins the queen to a discovered attack. Stockfish on our machine puts White at +5.02 or more, and the best Black can do is give up the queen for a knight.

The lesson is that 4.Qe2 attacks the knight on e4 and lines the queen up on the e-file against Black's king, so any move of the e5 knight can become a discovered check. If you play the Petrov as Black, play 3...d6 and make White retreat first. If you meet 3...Nxe4?, play 4.Qe2 and look for the discovered check. For the same method across other openings, see chess openings by rating.

The Cochrane Gambit: 4.Nxf7

The other sharp sideline is the Cochrane Gambit, 3.Nxe5 d6 4.Nxf7 Kxf7, where White gives a knight for two pawns and Black's king stays uncastled. Nick de Firmian called it speculative but entertaining in Modern Chess Openings, and he judges 5.Nc3 c5 as offering chances for both sides, citing Topalov against Kramnik at Linares in 1999. Our check is less romantic. After 4...Kxf7 Stockfish has White at -1.08, with 5.Nc3 as the best try, so the gambit costs White more than a pawn of evaluation. It suits a blitz surprise, not a search for an advantage.

The Stafford Gambit: 3.Nxe5 Nc6

The second dataset position is the Stafford Gambit, and it is a good example of why a quiet defence has a loud sideline. The path to it is not random. At 1400, the model gives 3.Nxe5 as White's likeliest move at 32.5 percent in the first position, and the same model's likeliest Black reply is 3...Nc6, at 50.5 percent. That probability exists only for the 1400 band, so we do not quote it for other ratings.

The Stafford is named for Joseph Stafford, a correspondence player who won a six move game with it in 1950, and Wikipedia gives an 1857 simultaneous exhibition as the earliest instance, where Howard Staunton lost as Black. It was rare in tournaments for decades. As of October 2026, Wikipedia describes it as a surprise weapon online in the 2020s, popularised by IM Eric Rosen's blitz games, and reports that in December 2024 Rosen beat GM Sergey Erenburg in 14 moves with it at the FIDE World Blitz Championship. We found no measured share of games, so we cannot say how often you will meet it. Wikipedia also says high level players consider it unsound in classical chess, rating the start position at roughly +1.5 to +2.0 for White.

The position after 3...Nc6 is the dataset's late position. Stockfish ranks 4.Nxc6 first at +1.70, then 4.d4 at +0.85 and 4.Nf3 at +0.09. 4.Nxf7 is -2.18.

The model finds the engine's move easily. 4.Nxc6 is predicted at 73.8 percent at 1000, climbing to 87.0 at 1600 and easing to 85.3 at 2200. The speculative 4.Nxf7 stays near 2.5 percent from 1000 to 1600. The danger for White is not move four but move five.

After 4.Nxc6 dxc6: play 5.d3, not 5.e5

This is the practical heart of the page. After 4.Nxc6 dxc6, Black has opened lines for the queen and the c8 bishop in exchange for a pawn. White has several good moves. Stockfish on our machine has 5.d3 at +1.58, 5.Qe2 at +1.53 and 5.Nc3 at +1.51, and after 4.Nxc6 dxc6 the position sits around +1.5 to +1.6 whichever of those White chooses.

5.e5 is the move that looks most natural, because it attacks the knight and takes space. It lets Black into a trap. Wikipedia describes the 1950 Lowens against Stafford game, where after 5.e5 Ne4 6.d3? Bc5! White resigned. We checked the position after 6.d3 Bc5 with Stockfish on our machine: White is at -3.21 even with the best defence, 7.Be3.

• 5.e5 Ne4 6.d3 Bc5. White is at -3.21 with 7.Be3 Bxe3 8.fxe3 Qh4+ as the best try.
• 7.dxe4?? Bxf2+. White is at -4.28. After 8.Kxf2 Qxd1 the queen is gone for a bishop.
• 5.d3 Bc5 6.Be2. White keeps +1.71, and 5.d3 is the move to learn. It defends the e4 pawn, avoids the overextension of 5.e5, and prepares Be2 and castling.

We only claim what we checked: the trap after 6.d3 is real, and 5.d3 is simpler. As Black, 5.e5 is the reply you hope for. As White, the plan is short: trade on c6, play d3, develop the bishop to e2, castle, and use the extra pawn.

What the numbers say about how to play it

Put the two tables together. Below 1400, the Petrov as Black is a very practical choice. You meet 3.Nc3 or 3.Nxe5 about equally, and about one game in five goes 3.Bc4, which you can answer with 3...Nxe4 for a small edge. When White plays 3.Nc3, 3...Nc6 heads for a Four Knights structure that needs little theory. At 1400 to 1800 you should know the main line 3.Nxe5 d6 4.Nf3 Nxe4, because 3.Nxe5 is likeliest. At 2000 and above, expect 3.Nxe5 about 40 to 48 percent of the time and 3.d4 more often than before.

For White, the choice depends on goals. 3.Nxe5 gives the engine's best score, +0.50, and if Black plays 3...Nc6 you have a clear plan. 3.d4 is +0.34, and our check of 3.d4 Nxe4 gives about +0.28 for White, a fully playable position. 3.Nc3 is +0.19 and invites the Four Knights at +0.08. Avoid 3.Bc4 if you want more than equality: at -0.58 it is by far the worst of the top five, more than half a pawn below 3.d3 at -0.08.

Plans matter more than move lists, the argument in why you should stop memorizing openings and learn plans. Black swaps off White's e4 pawn and builds a solid centre with ...d5. White tries to keep a small lead in development.

Is the Petrov a good opening?

Yes, for Black, with a caveat about what you want from a game. Black is fine in every main line we checked: the best White score in the main variation is +0.39, and in the 3.Nc3 line it is +0.08. Chess.com lists the cons as White keeping a slight development lead in open positions and a lot of theory. One chess.com forum commenter argues it is not drawish online at all, an opinion we cannot check.

If your goal is to beat a weaker opponent, the Petrov asks you to unbalance the position yourself. If your goal is a safe opening, it is hard to beat, though ChessBase argued in 2018 that it has many landmines. No statistics were given, so treat that as opinion.

Where other resources are better

This page is narrow on purpose. The dataset holds two positions, and Maia3 5m is a small model of human choice, so its percentages are predictions and can be wrong at the margins. The line checks are quick local runs, not a substitute for an opening database or a repertoire course. We have no game results by rating, so we cannot say how Black scores at your level, and we have no measured share for the Stafford. Wikipedia is the better reference for the full list of named variations, adopters and history, and chess.com is better for a short overview with a master level draw rate. Lichess has video lessons and an opening explorer for real game counts, and we could not reach its opening pages from our research tools. None of the pages we opened shows what White plays after 2...Nf6 by rating, which is the gap this one fills.

What to take from it

The engine prefers 3.Nxe5 at +0.50, but 3.Nc3 leads at 1000 and 1200 at 32.2 and 32.3 percent, and 3.Nxe5 takes over from 1400, reaching 47.6 percent at 2200. 3.Bc4, the worst of the top five at -0.58, is predicted 19.6 to 21.0 percent of the time from 1000 to 1800. In the Stafford, 4.Nxc6 is found 73.8 percent of the time at 1000 and 87.0 at 1600, and the move to remember afterwards is 5.d3, not 5.e5. As Black, play 3...d6, not 3...Nxe4. Maia3 is a model of human choice trained on online games, not a count of real games, and every percentage above should be read that way.