Rozman calls this the hardest chapter in the book to write. There are 400-to-500-page books devoted entirely to endgame theory, and he has only a few dozen pages to give you something useful. So this chapter trades depth for breadth: enough of each major endgame type — pawns, knights, bishops, rooks, queens — that you recognize the pattern at the board and know which plan to reach for, even though whole volumes exist on any one of these topics alone.
There's no strict cutoff for where the endgame begins, but here's a working definition. Each side starts the game with fifteen units that can be captured: eight pawns plus two knights, two bishops, two rooks, and a queen (the king is the one piece that can never be taken). Once roughly half of your seven non-pawn pieces have come off the board, together with a handful of pawns, you're no longer in a middlegame — you're in an endgame, with a different set of rules of thumb.
Some endgames are solved: with perfect play from both sides, the result — win, loss, or draw — is fixed in advance, regardless of how "good" or "bad" either side's remaining pieces look. King and pawn versus king is the classic example: there's no such thing as a slightly-better king and pawn ending, only a winning one, a drawing one, or a losing one. Other endgames are merely practical — not forced by theory, but winnable with the right technique and enough pressure applied over the board. Magnus Carlsen, the reigning world champion as this book went to print, built a reputation on exactly this kind of ending: he'll grind an opponent for five or six hours in a position that "should" hold with perfect defense, and sooner or later a human being sitting across the board makes the one inaccuracy that turns a draw into a loss.
Six factors decide the outcome of most endgames, and you'll see all six at work throughout this chapter.
| Factor | What it means at the board |
|---|---|
| Material | Who has more, and — crucially in an endgame — does the extra material actually matter with these particular pieces left on? |
| Activity | Whose king and remaining pieces reach the important squares first. |
| Weaknesses | Pawns or squares that can't be defended forever. |
| Zugzwang | Being forced to move on your turn when every legal move makes your position worse. |
| Domination | Pinning an enemy piece down with nowhere useful to go. |
| Passed pawns | A pawn no enemy pawn can stop from advancing — especially a dangerous one out on the wing. |
Chess vocabulary is a patchwork of borrowed words: zugzwang and zwischenzug are German, fianchetto and tempo are Italian, en passant and en prise are French.
Pawn endings are the foundation everything else in this chapter builds on, because so many other endgames get simplified down into one. Picture a bare king and pawn racing to promote against a lone defending king: if White is four squares from queening and it's White's move, nothing stops the pawn — it queens and delivers the king-and-queen mate you already know from Chapter Four. But if it's Black's move in that same position, Black's king can just barely get back in time to stop it, and the game is a draw.
That "just barely" has a name: the square of the pawn. Draw a square starting at the pawn, with sides equal to the number of moves the pawn needs to reach its promotion square. If the defending king can step inside that square on its own move, it catches the pawn no matter how far away it looked; if it can't get in, the pawn queens completely untouched, with no other pieces required at all.
Here's how the chase ends if Black gets there one tempo too late:
Once both kings have arrived at the pawn, everything hinges on one geometric detail: does the pawn stand in front of its own king, or has the king already gotten in front of its pawn? If the pawn leads the way, the position is a fortress draw — the defending king just keeps stepping back into the pawn's shadow, mirroring every advance no matter whose move it is.
If Black keeps repeating that same shadowing technique correctly, White can march the whole way up the board and still never make progress. Push it to its logical end and you get a position like this one, still with the pawn in front of the king — a reminder that "in front" endings are drawn no matter how advanced the pawn gets:
Flip the setup around — the attacking king already in front of its own pawn — and the very same face-off now favors the attacker instead of the defender. This is where outflanking comes in: whoever is forced to break the face-off has to give ground somehow, but not helplessly. Try to find it before you press Show solution.
Does that same outflanking idea work everywhere on the board? Try the pawn's most extreme case before you press Show solution.
Outflanking only works with the pawn on the b through g files. A rook pawn (a- or h-file) has no room on the far side to outflank into — the defending king can just plant itself in the corner in front of it, and the game is a draw no matter how well the attacker plays.
With more pawns on the board, an outside passed pawn — one on a file far from the rest of the action, that no enemy pawn can ever stop — changes the whole picture. Try to find it before you press Show solution.
In king and pawn endgames, king activity is worth more than it looks, and a passed pawn on the far edge of the board is worth more than a passed pawn in the center — even if it never queens, it can single-handedly drag the enemy king out of position.
King and knight can never checkmate a lone king, so the moment every pawn disappears from a knight ending, the result is automatically a draw — it doesn't matter who was "better" a move earlier. That single fact shapes the whole category, since the pieces themselves usually can't force anything on their own. Compare these two structures — nearly identical, but not quite. Try to spot the difference before you press Show solution on each.
That's the concept of two weaknesses: one side has both a distant, unstoppable pawn to worry about and a cluster of home pawns to protect, and a single king plus a single knight simply can't be in both places. Force your opponent to choose which weakness to abandon, and you win the other one.
A real game supplies a perfect illustration: Michael Adams versus Jonathan Speelman, 1995.
A "should be a draw" evaluation and an actual drawn result are two different things, though — this is exactly the kind of practical ending Carlsen has built a career on grinding out. A handful of moves later, White's knight lands a fork that both attacks Black's king and wins a pawn in one shot:
From there, White kept applying the same "two weaknesses" pressure until material changed hands again. Here's the position a little later, with White to move. Try to find it before you press Show solution.
A bishop crosses the whole board in a couple of moves; a knight needs several short hops to get anywhere far away. That speed gap can sometimes let one minor piece completely dominate the other. Try to find it before you press Show solution.
Black resigned in the position below, before White even had to prove it — the knight was still stuck, and the mopping-up was only a matter of time.
The reverse also happens, too — sometimes it's the knight that comes out ahead. Which piece is struggling here, and why? Try to find it before you press Show solution.
When most of the pawns — both sides' — end up sitting on the same color as a bishop, it becomes what's called a "bad" piece: no targets, and no way past its own pawn chain. A knight isn't restricted by color at all, so it can hop straight over that same wall, exactly as White's knight does here.
Knights can, in specific structures, outperform bishops in the endgame — but as a general rule, bishops are the stronger piece over the long run, since their speed and reach with pawns on both wings usually outweighs a knight's ability to dominate an occasional bad bishop. Chapter Fifteen returns to this comparison.
Bishop endings split into two very different worlds depending on whether the two bishops are the same color or opposite colors.
Opposite-colored bishops have a reputation for how stubbornly they draw, even when the material looks lopsided. Try to find it before you press Show solution.
Push it further — give White a second extra pawn, still connected on the kingside. Does that change anything? Try to find it before you press Show solution.
So what would it actually take to crack that fortress? Try to find it before you press Show solution.
So while scattered pawns are usually a weakness in the middlegame, in the endgame they can be a real asset — they split the defender's attention and give the opponent too many things to worry about at once. That's true in opposite-colored bishop endings, and it's just as true in same-colored bishop endings.
Magnus Carlsen demonstrated this technique against Alireza Firouzja in 2021, starting from a bishop ending that was both materially and structurally symmetrical.
Instead, over the following moves Carlsen quietly rearranged his own pawns onto dark squares — out of reach of Black's light-squared bishop — while freezing Firouzja's pawns in place on light squares, where his own bishop could snipe at them for free.
With that groundwork done, Carlsen gave up a pawn on purpose to free his king, which then marched straight toward Black's weakened queenside.
Carlsen turned a "shake hands" symmetrical ending into a full point by forcing his opponent into hard decisions on every move, then attacking pawns that were stuck on the wrong color while his own stayed safe on the right one.
Rook endings are the most common endgame in chess, and by reputation the most error-prone — entire manuals thousands of pages long are devoted to them. Here's a position where one side is up a pawn, but the real advantage runs deeper than the pawn count. Try to find it before you press Show solution.
Rook and pawn versus rook, though, is not automatically winning. The key defensive resource is the Philidor Position: the defending rook sits on the rank just in front of the pawn, blocking the attacking king from ever crossing it. The instant the pawn advances onto that rank, the rook's own view along it gets blocked — so it drops back to check from behind instead.
Does that same Philidor drawing technique hold up as the material gap grows? Try it before you press Show solution.
A grandmaster would agree to a draw in a Philidor-type position even three pawns to two, as long as the extra pawns are all on the same side and the defender knows the technique.
Same pawn count as Board 27 — three against two — but arranged differently. Does the result change? Try to find it before you press Show solution.
What actually wins rook endings is a pawn majority on the other wing from where the action is — one side can't defend the loose pawn over there and hold the kingside together at the same time.
Pawns bunched on one side of the board tend toward a draw in rook endings; pawns scattered across both wings tend toward a win for whoever has the extra one. Keep that contrast in mind — it's the single most useful rule of thumb in this section.
Queen versus a lone pawn is fully solved, and — with one specific exception you need to memorize — it always wins for the side with the queen. The method: get the queen in front of the pawn so it can't move, then walk your own king all the way up the board to help capture it and finish with the king-and-queen mating technique from Chapter Four.
The trickier case is when the pawn is one square from queening and already defended by its own king. The technique still wins outright — but only on the b, d, e, and g files.
On the a and h files, though, that same checking method runs into a problem. White has just forced Black's king in front of its own pawn — the natural next step looks like bringing the White king in closer to help. Try to find out whether that's actually safe before you press Show solution.
The c- and f-pawns have their own quieter version of the same trap. Try to find it before you press Show solution.
| Pawn's file | Queen vs. lone pawn, king defending it on the 7th/2nd rank |
|---|---|
| b, d, e, g | Winning — the checking technique drives the king in front, then the attacking king walks in. |
| a, h | Drawn — driving the king in leaves no room; taking the pawn or improving is stalemate. |
| c, f | Drawn in the critical setup — the defending king slips to the corner and lets the pawn go, since capturing it is stalemate. |
Whenever you're mopping up a lone pawn with your queen and the enemy king is tucked in a corner, stop and check for stalemate before you take the pawn. It costs nothing to bring your own king a little closer first.
Queen endgames with pawns on both sides play very differently from everything else in this chapter. Material imbalances matter less, because a lone queen alone can manufacture serious threats out of nothing, and kings are usually tucked away in a safe corner rather than centralized the way they are in pawn or rook endings — an exposed king invites perpetual check, an unstoppable string of checks that forces a draw by repetition no matter how far ahead you are on material.
Being clearly winning on material in a queen ending is not the same as being safe. If your king isn't tucked somewhere the enemy queen can't reach, a material lead can evaporate into a perpetual-check draw.
Endgame evaluation always comes back to the same short list: material, activity, weaknesses, zugzwang, domination, and passed pawns. Pawn endings are the base case everything else reduces to — know the square of the pawn, the opposition, and outflanking, and you can convert or hold a huge share of the endgames you'll actually face. Just remember the one hard limit: outflanking fails on the rook files, because there's no room to slide around the defender.
King and knight can never mate a bare king, so knight endings are pawn endings in disguise — the outside passed pawn and the "two weaknesses" it creates are almost always the deciding factor, as in the Adams–Speelman game. Bishops generally outperform knights in the endgame thanks to their speed across the whole board, though a knight can dominate a "bad" bishop that's stuck behind its own pawns on the wrong color, and a bishop can just as easily trap a knight that has nowhere safe to jump.
Opposite-colored bishop endings are drawish even a pawn or two down, since the defending bishop and king only need to hold one front — but scattered pawns on both wings overwhelm that defense by opening a second front. Same-colored bishop endings hinge on activity and zugzwang much like pawn endings do, as Carlsen showed against Firouzja by rearranging pawns onto the "safe" color before breaking through.
Rook endings are the most complex and the most common of all — cutting off the enemy king is the attacker's main tool, and the Philidor Position is the defender's; pawns bunched on one wing tend to draw, pawns split across both wings tend to win. Queen versus a lone pawn is winning except in specific stalemate traps on the a, c, f, and h files, and queen-and-pawn endings turn on king safety and the threat of perpetual check as much as on the raw pawn count.
A solved endgame has a known result — win, draw, or loss — with best play from both sides; king and pawn versus king is a classic example. A practical endgame isn't forced in that sense, but is winnable with the right technique and pressure; the defender might hold with perfect play, but real opponents often don't find it, which is exactly the kind of ending Magnus Carlsen has built his reputation grinding out.
Count how many moves the pawn needs to reach its promotion square, and draw a square of that many files and ranks starting from the pawn's square. If the defending king can step inside that square on its move, it catches the pawn unaided; if it's outside and can't get in, the pawn queens.
Opposition matters whenever the kings are close to the pawn's path: the side forced to move away from a face-off usually has to give ground, letting the other king advance or outflank. It stops mattering once one king is comfortably in front of its own pawn with room to outflank on either side, or once the pawn is a rook pawn and the defending king has already reached the queening corner — there's simply no room to outflank into.
Pushing straight ahead runs into Black's king again. Sliding to the far side (outflanking) claims space Black's king can't contest without abandoning the pawn's path entirely — Black can only shadow from a distance, which is exactly what lets White's king and pawn walk in together for the win.
A lone king and knight cannot force checkmate against a bare king, so if all the pawns get traded off in a knight ending, it's a draw regardless of any earlier piece activity. That's why the winning plan in knight endings is almost always about pawns — especially an outside passed pawn creating "two weaknesses" that one king and one knight can't both cover, as in the Adams–Speelman game (Board 11).
A bishop is bad when its own side's pawns sit on its color, blocking its diagonals and leaving it with few useful squares. A knight can exploit this by hopping to both colors and both sides of the board while the bishop stays boxed in with nothing to attack, as in Board 15 and 16, where every White pawn sits on a dark square, completely shutting out Black's light-squared bishop.
The defending bishop can sit on a square of its own color directly in a pawn's path, or sacrifice itself for it, and a king plus bishop alone cannot checkmate a bare king. Even two connected extra pawns often still lose to one well-timed blockading move (Board 18). What breaks it is scattered pawns on both wings (Board 19) — the defender can't rush to two fronts with only one king and one bishop.
Zugzwang is being forced to move on your turn when every legal move worsens your position. In Board 20, Black's bishop is the only piece guarding two pawns at once; moving it drops one, and moving the king instead lets White's king infiltrate the other side. There's no move that doesn't lose something.
It's the standard drawing technique in rook and pawn versus rook: the defending rook sits on the rank just ahead of the pawn, stopping the attacking king from crossing it. Once the pawn itself reaches that rank, it blocks the rook's own view along it, so the rook can no longer control the far side — it switches to checking the attacking king from behind instead, holding the draw by repetition.
In Board 27 all five pawns sit on the same wing, so the Philidor technique holds regardless of the extra pawn. In Board 28, White's extra pawn is alone on the other wing, creating the same "two weaknesses" problem as in the knight endings: Black's king and rook can't defend the kingside and stop the lone pawn at the same time.
Almost always — the winning technique works on the b, d, e, and g files. On the a and h files, and in the specific king setup shown for the c- and f-pawns, the same checking method runs the defending king into a stalemate instead, so those particular configurations are draws.
A lone queen generates enough threats that material imbalances matter less than elsewhere, and an exposed king is a serious liability because the opponent's queen can often force a perpetual check. Tal was two pawns up against Botvinnik but had to spend real effort tucking his own king to safety before that material edge meant anything — in a rook ending the same material would have been a trivial, immediate win.