Scotch Game: What Every Rating Actually Plays After 4.Nxd4

The Scotch Game is the quickest way to open the centre after 1.e4 e5 2.Nf3 Nc6: 3.d4, and White offers a pawn trade at once. Most guides then follow the master games and tell you Black replies 3...exd4 4.Nxd4 and chooses between 4...Nf6 and 4...Bc5. For a player below 2000 that is not what usually happens. We ran the maia3-5m checkpoint, see how Maia chess actually works, on the position after 4.Nxd4 at seven rating bands from 1000 to 2200, the same method behind our chess openings by rating dataset, and graded every reply with Stockfish 18, see what the Stockfish chess engine is. From 1000 to 1800 the likeliest reply is 4...Nxd4, trading knights so that White's queen recaptures in the centre. It is also the worst of the three replies the guides discuss. Maia3 predicts what a player at a given rating is likely to choose. It is not a count of real games.

One convention before the numbers. Every evaluation on this page is from White's side, so +0.16 means White is better by 0.16 of a pawn and -0.17 means Black is slightly better. Our table figures are Stockfish's score for the position after Black's move, restated that way.

A short history of a long-named opening

The Scotch begins 1.e4 e5 2.Nf3 Nc6 3.d4 and sits in ECO codes C44 and C45. Wikipedia says it got its name from a correspondence match between Edinburgh and London in 1824. The same page says Ercole del Rio was the first author to mention what is now called the Scotch, in a 1750 treatise, and that the opening declined by 1900 because it was thought to release the central tension too early and let Black equalise.

The revival is better attested. Wikipedia says Garry Kasparov led a revival in the late 20th century, arguing that it caused Black lasting strategic problems while avoiding the heavily analysed Ruy Lopez. Search summaries of ChessBase pages place that in his 1990 world championship match against Karpov, in Games 14 and 16, and call it his first Scotch. We did not verify the results from a game page. Edward Winter's page shows those games reached the Mieses structure and notes that the same lines had been argued by Steinitz and Zukertort in the 1880s. Wikipedia adds that Jan Timman and Sergei Rublevsky have advocated it, and that Vassily Ivanchuk and Magnus Carlsen revived the Potter Variation in the 2000s.

Black's third move: 3...exd4 is not automatic

Wikipedia says Black almost always replies 3...exd4, and at the top of the rating scale that holds. For everyone else the picture is looser. Stockfish scores 3...exd4 as best at +0.16. The alternatives are 3...Nxd4 at +0.64, 3...d5 at +0.88, 3...d6 at +0.92 and 3...Nf6 at +1.07. Here is how often Maia3 predicts the engine's choice.

At 1000, fewer than half of the predicted replies are 3...exd4. 3...d6, the quiet defence of the e5 pawn, is the main alternative, at 16.7 percent. It falls away steadily as ratings rise. 3...Nf6 also appears in the top five at 1000 and 1200 with 8.4 and 6.1 percent, and 3...f6 with 8.3 and 5.3. The honest reading is that a 1000 rated Black is close to a coin flip between taking on d4 and doing something else, and that White's central pawn move works as a filter on knowledge. By 2200 it is 95.6 percent.

After 4.Nxd4, the reply the guides do not rank first

This is the position that matters. After 3...exd4 4.Nxd4, Stockfish's ordering of Black's fourth move is 4...Nf6 at +0.16, 4...Bc5 at +0.27, 4...Qf6 at +0.33, 4...Bb4+ at +0.55, 4...d5 at +0.67, 4...d6 at +0.69 and 4...Nxd4 at +0.71. In the table, "out" means the move was outside Maia's top five at that band. It does not mean zero probability.

Read the first row. 4...Nxd4 is predicted at 57.8 percent at 1000, 58.4 at 1200, 54.0 at 1400, 45.4 at 1600 and 33.3 at 1800. Over the same range 4...Nf6 sits between 10.8 and 13.6 percent, and it does not reach 21.0 until 2000. 4...Bc5 climbs steadily and takes the lead at 2000 and 2200, at 37.0 and 39.0 percent. The likeliest move is then 11 centipawns from the best, where from 1000 to 1800 it is 55 centipawns away. At 2200, 4...Nf6 is close behind at 35.8 percent. 4...Qh4, the Steinitz, enters the top five only at 2200, with 4.5 percent, and the engine scores it +0.85.

The gap is the story. 365chess ranks 4...Nf6 first at master level, as of October 2026, and the forum treats 4...Nf6 and 4...Bc5 as the choice. pfren writes that "I have always played 4...Nf6 and I always found the resulting positions easier to handle than after 4...Bc5." That is good advice, and it is also advice that most 1000 to 1800 players do not follow. No page we read tells club players that the move they are likeliest to meet is a third option.

What to play as White against 4...Nxd4

You recapture with the queen: 4...Nxd4 5.Qxd4. 3...Nxd4 4.Nxd4 exd4 5.Qxd4 reaches the same position. Stockfish scores it about +0.7: +0.71 at depth 20 in our dataset, and +0.68 and +0.74 in two depth 22 runs. That is a clear small edge. The engine's line for Black is 5...Qf6 6.Qc4 c6 7.Nc3 b5 8.Qd3.

The plan follows from that. Your queen is in the centre but not in danger, so develop the queenside knight, bring the light bishop out, castle, and treat the lead in development as the asset. Black has given up the knight pair and has a queen that wants to hit your queen with ...Qf6, so do not be surprised by it. If you want help turning this kind of plan into habit, why you should stop memorizing openings and learn plans covers the method.

The comparison with the other replies matters if you are White. Against 4...Nf6 or 4...Bc5 you face roughly +0.16 to +0.27, so a quiet edge of about a quarter pawn at best. Against 4...Nxd4 you start with about +0.7. The reply that is likeliest at your rating is the one that gives you the most.

The named lines, with engine numbers

Move orders and names follow Wikipedia's list. Every score below is from our local Stockfish 18 run at depth 22, from White's side.

The Classical Variation, 4...Bc5, is the straightforward developing reply. A typical line is 5.Be3 Qf6 6.c3 Nge7 7.Bc4, which scores +0.26 with 7...Ne5 as Black's best move. The sharper alternative, 5.Nxc6 Qf6 6.Qd2 dxc6 7.Nc3, is +0.09, nearly level. Chess.com puts the Classical at White 37 percent, draws 38 and Black 25, as of October 2026, but the page gives no pool or rating band and its labels were unclear in our read, so hold the figure loosely.

The Potter Variation is 4...Bc5 5.Nb3 Bb6. It scores +0.26, with 6.Qe2 the engine's choice for White. Wikipedia credits Ivanchuk and Carlsen with reviving it in the 2000s.

The Schmidt Variation is 4...Nf6 5.Nxc6 bxc6, the doubled pawns line. The Mieses Variation grows out of it with 6.e5 Qe7 7.Qe2 Nd5 8.c4 Ba6. In our check the position after 8.c4 is +0.00 for White, with 8...Ba6 as the engine's choice. The honest summary is that the main line is level. Chess.com puts the Schmidt at White 36 percent, draws 33 and Black 31, as of October 2026, again with no pool stated. Ethan_Brollier's forum claim that "Black just equalizes instantly" is a little strong, since 4.Nxd4 Nf6 is still +0.16 for White, but maafernan's more careful view fits the engine better: "not easy to counter," yet level with accurate play.

The Scotch Four Knights is the structure after 4...Nf6 5.Nc3 Bb4. Our check of 6.Nxc6 bxc6 7.Bd3 d5 gives +0.16 for White, with 8.exd5 the engine's pick. Wikipedia notes that 4...Nf6 5.Nc3 transposes to the Scotch Variation of the Four Knights Game. For White players who want less theory across the 4...Nf6 lines, Chess.com's GM Gserper recommends a quieter Bd3 setup instead of the main lines, saying the heavy theory of the main lines punishes the unprepared, and makes the useful point that at club level "instead of playing Kasparov or Anand, you will still play your 1800 rated club-mate."

The Steinitz Variation, 4...Qh4, attacks the e4 pawn at once. After 5.Nc3 Bb4 6.Be2, the capture 6...Qxe4 is the engine's own best move, so it is not a blunder. The position is +0.70 for White, and +0.91 after 7.Nb5, which is a clear edge, not a refutation. A search summary of Wikipedia's Steinitz page says the idea forces the win of a pawn at a major cost of time, and gives White's most successful line, as of 2005, as 7.Nb5 Bxc3+ 8.bxc3 Kd8 9.O-O. Black's king has lost castling rights and Black is a pawn up, and the engine still prefers White.

The gambits: engine and practical scores disagree

The Scotch Gambit is 4.Bc4, and a search summary describes it as leading to the most tactical and sharp positions in the whole Scotch complex. Stockfish scores it +0.01, which is level, with 4...Bb4+ as its answer. That is 15 centipawns less than the +0.16 of 4.Nxd4, which is not a failing. illegalthingy's forum line that the Gambit is "the only thing playable" is wrong by that measure, since the quiet 4.Nxd4 keeps a larger edge.

The sharpness is real, but do not expect free material. One line a player might hope for runs 4...Bc5 5.Ng5 Nh6 6.Nxf7 Nxf7 7.Bxf7+ Kxf7 8.Qh5+ g6 9.Qxc5. White ends a pawn up on paper, but our check scores the position +0.00, so the extra pawn is worth nothing. We found no sourced named trap with a move list for the Scotch, and we are not presenting one.

The Goring Gambit, 4.c3, is the one place where the engine and the practical results disagree. Stockfish 18 gives it -0.17 for White and -0.14 after 4...dxc3 5.Nxc3, a small deficit. Chess.com, as of October 2026, shows White 44 percent, draws 26 and Black 32, the best White score among the lines it lists, with no pool or rating band. The same position also arises from 1.e4 e5 2.d4 exd4 3.c3 dxc3 4.Nxc3 Nc6 5.Nf3, the one-pawn line covered in the Danish Gambit by rating. A gambit that the engine rates a shade worse can still score well against players who do not know it, which is the usual story for gambits at club level. It does not make 4.c3 better than 4.Nxd4. It makes it a reasonable surprise weapon if you know the lines.

Scotch or Italian, and how much to learn

Chess.com calls the Scotch one of the best alternatives to the Ruy Lopez and says that avoiding the heavy theory of that opening is especially advantageous for beginners. 365chess puts 3...exd4 at 41,468 games, as of October 2026, with White 40.9 percent, draws 29 and Black 30.2, though it does not state the database or the rating band. 365chess also lists the drawbacks: releasing the central tension early can give Black counterplay, and the extra tempo spent recapturing on d4 can disrupt development. On how it compares with the Italian, Ethan_Brollier, in a chess.com Ruy Lopez thread, recommends "the Scotch or the Italian" and suggests avoiding theory until about 1050.

If you want to compare it with the quieter 3.Bc4 on the same dataset, our Italian Game page does the same exercise, and the Ruy Lopez page covers 3.Bb5, where the likeliest reply also changes with rating. Our data suggests a simple division of study time. Learn 5.Qxd4 first, since it is what you are likeliest to meet below 2000. Add 4...Bc5 next, and keep 4...Nf6 for last unless you play above 1800.

Where other pages are better

This page is narrow on purpose. Our data covers Black's third and fourth moves after 3.d4 at seven ratings. Everything past those points comes from a handful of local Stockfish checks and from other writers, and it is not a database. We do not measure White's fifth move, and the Lichess explorer, which would give rating filtered results, was unreachable, so we have no club level win rates. Wikipedia is the stronger reference for the history and the full list of named variations, and chess.com and 365chess are better for browsing lines. None splits Black's fourth move by rating, which is the gap this page fills, and none shows an engine score for every reply.

What to take from it

The guides teach 4...Nf6 and 4...Bc5, and at 2000 and above the data agrees about 4...Bc5. Below that, you are most likely to meet 4...Nxd4 and the early knight trade. It costs Black 55 centipawns against the engine's choice, and it gives White the simplest plan in the opening. As White, learn 5.Qxd4 first. As Black, 4...Nf6 is the engine's choice, and the Mieses line it leads to is level. Maia3 is a model of human choice built from the maia3-5m checkpoint, not a count of real games, and every percentage above should be read that way.