Torre Attack: a practical guide to 3.Bg5, and which defences the system actually beats

Short answer

The Torre Attack builds the same structure every game — d4, Nf3, Bg5, e3, Nbd2, Bd3 and c3 — and sidesteps the main lines of the Queen's Gambit. This guide shows the system move by move, measures Black's replies on Stockfish 18 at depth 26, and says where the system pays and where it does not: against 1...d5 and 3...Ne4 the engine favours Black, while against the ...e6 and ...g6 structures White comes out slightly better. It includes the critical b2-pawn line, where Black's queen may capture but must not stay.

The essentials in a few lines

• After 1.d4 d5 2.Nf3 Nf6 3.Bg5, Stockfish 18 gives −0.12 at depth 26 — an edge for Black. The reason is 3...Ne4, which attacks the bishop and forces White to choose early between trading, retreating and losing time.
• In the same position, the London System with 3.Bf4 measures +0.15. That is this guide's most useful comparison: against ...d5, the bishop on f4 is worth about a quarter of a pawn more than on g5, because it cannot be hit by ...Ne4.
• Against the ...e6 structures the Torre is a different opening. After 3...e6 4.e3 Be7 5.Nbd2 O-O 6.Bd3 c5 7.c3 Nbd7, White stands at +0.42 with 8.Ne5 — the manoeuvre that gives the whole system its point.
• The critical line is not a trap: after 4...Qb6 5.Nbd2 Qxb2 6.Bd3, taking the pawn is correct, and the queen has two good follow-ups: 6...Qa3 (+0.14) and 6...d5 (+0.15) at depth 26. What does not work is pressing on: with the greedy 6...Qc3 7.O-O it rises to +0.62, and after 7...cxd4 it reaches +1.99, with 8.Bxf6! as the route.
• Against Black's fianchetto the e4-e5 push does not work: after 3...Bg7 4.Nbd2 O-O 5.e4 d5 6.e5 Ne4 the engine gives −0.21 for White. The road is the modest e3, with the bishop dropping back to f4 when attacked.

Three moves and a system

The name comes from the Mexican player Carlos Torre Repetto, and the opening belongs to the same family as the London System, the Colle System and the Trompowsky Attack: all of them settle the dark-squared bishop first and worry about theory later. What separates them is a single square — f4, g5 or inside the chain — and that is a bigger difference than it looks, as this guide's measurements show.

The practical appeal is the usual one for systems: a single body of knowledge against several defences. A player who opens 1.d4 and does not want to study the Queen's Gambit Declined, the Slav, the Nimzo-Indian and the King's Indian separately builds the Torre and plays the same seven piece moves against all of them. What changes from one defence to the next is the middlegame plan, not the order of development.

What this guide adds is the part usually missing: the system does not pay equally against every defence, and the difference is measurable. Against the ...e6 and ...g6 structures the g5 bishop finds a target and White comes out better; against an early 1...d5 the bishop becomes the target, and the engine prefers Black. Anyone adopting the Torre needs to know which of the two cases they are in before move four.

Against 1...d5 the problem is called 3...Ne4

It is worth understanding why this knight differs from any other knight on e4. Without a black pawn on d5, the advanced knight is usually a temporary guest: White plays Nbd2 followed by f3 or c4, and it goes back. With d5 played, the e4 square has permanent pawn support, and it is the white bishop that has to decide — 4.Bf4 hands over the initiative, 4.Bh4 leaves the piece on the wrong diagonal (−0.31), and 4.Be3, the engine's preference, admits the bishop came out for nothing and measures −0.14.

The direct comparison with the London System makes the cost visible. In exactly the same position, with 3.Bf4 instead of 3.Bg5, the engine gives +0.15 — a difference of nearly three tenths of a pawn, explained in one sentence: on f4 the bishop cannot be attacked by ...Ne4, so it loses no time. Worth recording too is the Trompowsky Attack (1.d4 Nf6 2.Bg5, before Nf3), which measures 0.00: same square for the bishop, one tempo more to react, intermediate result.

The practical conclusion is not to abandon the Torre but to order the repertoire: against an immediate 1...d5, the version with the bishop on f4 is the one the engine endorses; the bishop on g5 is for when Black plays ...e6 or ...g6. Anyone who wants to play 3.Bg5 against ...d5 anyway should do so knowing there is no opening edge on offer, and that the position after 3...Ne4 is playable but slightly worse.

Where the system pays: the ...e6 structures

The move 8.Ne5 is the heart of the system, and the reason the Torre is not merely passive development. The knight takes the outpost supported by d4 and by the f-pawn once f2-f4 is played; if Black trades it off with 8...Nxe5 9.dxe5, the e5 pawn evicts the f6 knight and opens the d3 bishop's diagonal towards h7. If Black does not trade, the knight stays. Both branches favour the side that built the system.

The pin on g5 has a concrete function in this structure: it makes ...c5-c4 and ...e6-e5 harder, the two breaks with which Black normally frees the position. As long as the f6 knight is pinned against the queen, every black break needs one extra preparatory move, and that is the tempo White uses to complete the kingside build-up. It is the Colle-Zukertort mechanism with the bishop already outside the chain.

The size of the edge deserves an honest note: +0.42 is a small opening advantage, comparable to any solid white system, not a refutation. The value of the Torre in these positions is not in the number but in the fact that it is obtained with no memorised theory — Black, arriving from defences with twenty catalogued moves, reaches move eight in a structure that was not part of the preparation.

• White's order is fixed: d4, Nf3, Bg5, e3, Nbd2, Bd3, c3, O-O and then Ne5.
• Against ...h6, trading with Bxf6 is normal and not a concession: it gives up the bishop pair but gains time and structure.
• The break to prevent is ...e6-e5; the pin on g5 is what delays it.
• If Black plays ...Ne4 in this structure, the answer is the trade on e7 or the retreat to f4, never keeping the bishop on h4.

The b2 pawn, and the queen that cannot stay

This is the line that decides whether the Torre is playable against 2...e6, and it is usually presented wrongly from both sides. White players are told the capture on b2 is greedy and loses; Black players are told the pawn is free. The measurement says something else: the pawn can be taken, and two continuations hold the balance — the retreat 6...Qa3 and the developing 6...d5 — but neither is the move intuition suggests, which is to keep the queen going.

The technical detail is 6.Bd3, a move that defends nothing and prepares everything. From there the a1 rook gets the b-file with Rb1 and short castling connects the pieces; the black queen on b2 runs out of useful squares, and every move she spends looking for one is a developing move for White. After 6...Qa3 or 6...d5 Black hands back the initiative and the position returns to balance; after 6...Qc3 Black is half a pawn worse for not having stopped in time.

The 8.Bxf6 of the punishing line deserves a note because it runs against the instinct of anyone playing the system: the g5 bishop, the piece the whole opening was built around, is exactly what gets traded to open the file and trap the queen. It works as a general reminder about systems — the piece that gives the opening its name is not the piece that must be preserved, and the engine has no attachment to it whatsoever.

Against the fianchetto, and the push that is not there

The contrast between the two lines summarises the whole system. With e3, White keeps the centre closed, the g7 bishop bites on granite and the black knight that went to e4 ends up having to come back — White's structure is modest but has no targets. With e4-e5, the long diagonal opens for the black bishop at exactly the moment White's pieces are still uncoordinated, and the knight on e4 gains permanent support.

Notice that ...Ne4 appears in all three sections of this guide with different results: against ...d5 it is strong (the knight has pawn support), against ...e6 it is neutral (White trades or retreats with tempo), and against ...g6 it is premature (the knight is evicted and the tempo returns to White). Recognising which of the three is on the board is, in practice, all the specific knowledge the Torre Attack demands.

To complete the comparison with the London System on this terrain: against Black's fianchetto both systems produce almost identical positions — +0.19 for the London and +0.13 for the Torre in the lines measured — and the choice between f4 and g5 becomes a matter of taste rather than evaluation. It is against 1...d5 that the difference shows, which is why many practical repertoires combine the two — London against ...d5, Torre against ...e6 and ...g6.

Using the Torre in a repertoire

The selection rule fits into a question asked on move two: is Black's pawn already on d5? If it is, the bishop goes to f4 and the game is a London System. If Black has not committed yet, or has played ...e6 or ...g6, the bishop goes to g5 and the game is a Torre. It is a decision made before any theory, and it solves the problem the measurements identified without extra study.

Training a system differs from training a theoretical opening, and this is worth insisting on because it is what makes it economical: there are no lines to memorise, there are three positions to recognise — the ...e6 structure, the ...g6 structure and the b2-pawn line. Reviewing your own games with an eye on which of the three appeared, and on whether the knight ever reached e5, teaches more than any list of variations.

The limitation to declare is the one every system shares: against someone who knows it, the Torre holds no surprise. Black knows the knight is heading for e5, knows the pin interferes with ...e5, and plays ...h6 at the right moment to force the bishop to commit. The +0.42 is real, but it is the edge of the player with a plan over the player who has not studied — not the edge of a better opening.

Frequently asked questions

Openings mentioned in this guide

• Torre Attack (A46)
• Trompowsky Attack (A45)
• London System (D02)
• Colle System (D05)
• King's Indian Defense (E60)