Anatoly Karpov replayed from memory, without a board, games played twenty years earlier. Move by move. Variation by variation. Journalists tested him several times: he did not slip.

Yet if you had asked Karpov to memorise a list of thirty unrelated words he would have done exactly like you: forgotten half within an hour.

This paradox is not a mystery reserved for Soviet champions. It is one of the twentieth century's most important findings in cognitive psychology, and it has a direct consequence for how you train. If you have played chess for years and plateaued, there is a good chance this explanation clarifies why.

The uncomfortable paradox

Here is the world's simplest experiment to test chess memory. You show a middlegame position for five seconds. You hide the board. You ask the player to reconstruct it from memory.

A novice places three or four pieces correctly out of thirty-two. A grandmaster places twenty-eight or twenty-nine.

Now you take the same pieces, scatter them randomly on the board, any piece on any square, and repeat. Result: grandmaster and novice perform about the same. Both place three or four.

The master's advantage vanishes completely.

This result, published in 1973 by William Chase and Herbert Simon in Cognitive Psychology, overturned the idea that chess champions had photographic memory or off-scale intelligence. What they have is more precise: a library.

Chase and Simon (1973): the experiment that changed everything

Their protocol is simple. Players at three levels (novices, intermediates, masters) see board positions for five seconds, then reconstruct what they saw. Positions come from real tournament games.

On real positions masters crush everyone. On random positions nobody beats anyone else.

Chase WG & Simon HA. (1973). Perception in chess. Cognitive Psychology, 4(1), 55-81.

Their conclusion redefines what memory means in expertise: masters do not memorise individual pieces. They memorise chunks, clusters of pieces forming recognisable motifs. A pawn on f2, g2, h2, king on g1: not four objects. One chunk: "intact kingside castling."

A novice sees thirty-two objects on the board. A master sees five to seven functional groups. Information compression draws on long-term memory, not on superior brute memorisation capacity.

What this implies: grandmasters' spectacular memory is not innate talent. It is the result of thousands of hours exposed to meaningful configurations. Not a gift. A library built move by move, game by game, year after year.

Before Chase and Simon: de Groot had already seen it all

Chase and Simon did not appear in a vacuum. They build on a pioneering thesis by Adriaan de Groot, Dutch psychologist and master-strength player, published in English as Thought and Choice in Chess in 1965.

de Groot asked players at different levels to think aloud facing tournament positions. His main discovery is counter-intuitive: what distinguishes grandmasters is not how far they calculate. Masters do not calculate deeper than club players. What differs is the quality of the first move considered.

de Groot AD. (1965). Thought and Choice in Chess. The Hague: Mouton.

A master almost immediately sees strong moves. A novice wastes time on weak moves before (sometimes) finding the right idea. The master is not faster because they calculate less; they are faster because long-term memory already presents relevant hypotheses. The library works before conscious reflection begins.

Template theory: what Chase and Simon had not yet seen

Twenty years after the original experiment, Fernand Gobet and Simon refined the model. Chunks explain how players memorise a standard position quickly. They are not enough to explain how real grandmasters, those who play twenty-board simuls or replay a twenty-year-old game like Karpov, manage far higher complexity.

Gobet and Simon proposed templates: cognitive structures wider than chunks, organised around a fixed core (characteristic pieces of an opening or typical plan) and variable "slots" the player adapts to the actual position.

Gobet F & Simon HA. (1996). Templates in chess memory: A mechanism for recalling several boards. Cognitive Psychology, 31(1), 1-40.

A grandmaster does not encode 100,000 distinct positions. They encode a few thousand flexible templates, each applicable to many slightly different positions. Their memory advantage is not quantity: it is organisation.

For you as a club player: every time you play the same pawn structure, reach the same endgame, recognise the same mating pattern, you reinforce a template. That is why openings are not only theory: they build reusable cognitive structures. And understanding why a plan works in a given structure beats memorising variations by heart a hundred times over.

Working memory: the variable everyone confuses with talent

George Miller showed in 1956 that working memory capacity (how much information you can manipulate at once) is about seven items for most people. Plus or minus two. Not fifty. Not one hundred. Seven.

Chess players do not "break" that limit. What they do is compress each item into an increasingly rich chunk. A novice allocates one working-memory slot per piece. A master allocates one slot per functional structure that may contain five to eight pieces. Same raw capacity, radically different yield.

That principle, which Anders Ericsson, theorist of deliberate practice, documented across domains, holds here too. An experienced musician does not retain a score note by note; they retain phrase by phrase, section by section. A grandmaster does not calculate square by square; they think in structures.

Neuroimaging studies confirm the shift: experienced players activate long-term memory areas (hippocampus, dorsolateral prefrontal cortex) more and "pure" working-memory zones less when recognising positions. Their brain does not store more; it sorts better. That evolution of brain geography under practice is described in detail in the article on chess and the brain.

The Magnus Carlsen case: the library that plays for him

In 2004 Carlsen was thirteen. A journalist showed him a famous game position. He recognised it instantly, naming the player, year, and tournament. Not because he has photographic memory (he does not remember relatives' phone numbers better than you do). Because he had already seen that structure dozens of times, encoded as a template: fixed core (pawn formation, king placement), variable slots (positional details).

Gobet and Simon formalised that mechanism in 1996. Carlsen probably does not know it by that name. But he is the best-documented living demonstration: roughly 30,000 hours of reflective practice built a template library so dense that recognising a position feels like reading a familiar word. No spelling out. No calculation. Instant retrieval from long-term memory.

What this means for you: Carlsen does not memorise more variations than his rivals. He memorises them better, in flexible structures, not lists. That is accessible. Not quickly, but it is accessible.

What this changes for your training

If chess memory rests on accumulated patterns rather than innate capacity, the training implication is immediate:

What builds chunks and templates:

  • Deep analysis of your own games (reinforcing motifs you already met in real positions)
  • Study of typical endgames (king and pawn, rook and pawn: repeated positions that encode deeply)
  • Review of annotated grandmaster games (exposure to high-level templates with grasp of the plan)
  • Solving tactical puzzles while understanding why the move wins, not only which move

What builds little or none:

  • Playing fifty blitz games without analysing any. You see patterns but fail to consolidate them in long-term memory. Chunks do not anchor in the rush
  • Memorising opening lines without understanding the underlying plan. You store lists, not templates. Once your opponent deviates by one square the list becomes useless

Ericsson often cited chess studies to illustrate this: gains in domain-specific memory come from reflective practice, not raw volume. The distinction between "playing" and "training" is exactly there.

The same mechanism explains why some cognitive profiles, those who naturally recognise visual patterns, find chess especially suited to their information-processing style. That is explored in depth in the article on chess and autism.

In summary

Concept What it means in practice
Chunks (Chase & Simon, 1973) A master memorises functional groups of pieces, not individual pieces. The edge disappears on random positions.
Templates (Gobet & Simon, 1996) Broad, flexible cognitive structures handle complex positions without encoding each configuration separately.
Working memory Raw capacity is similar for everyone. Strong players pack more information per slot thanks to chunks.
de Groot (1965) Masters do not calculate farther; they see strong moves first because long-term memory suggests good hypotheses.
Practical takeaway Effective training targets pattern construction through reflective practice, not volume of games played.

Key takeaways

  • Grandmasters do not have globally superior memory: their scores on standard tests (word lists, digits) are ordinary
  • Their edge is domain-specific: they memorise "chunks", functional groups of pieces, not individual pieces (Chase & Simon, 1973)
  • Gobet and Simon (1996) refined the model with "templates": flexible cognitive structures that handle very complex positions
  • Practical consequence: playing fifty blitz games without analysis barely builds those structures. Effective training is reflective, not volumetric

Sources and references

  • Chase, W. G., & Simon, H. A. (1973). - Perception in chess. Cognitive Psychology, 4(1), 55-81. (Founding experiment: masters memorise functional groups of pieces, not individual pieces. On random positions their advantage vanishes.)
  • de Groot, A. D. (1965). - Thought and Choice in Chess. The Hague: Mouton. (Pioneering chess expertise psychology: grandmasters see strong moves immediately because long-term memory already offers relevant hypotheses.)
  • Gobet, F., & Simon, H. A. (1996). - Templates in chess memory: A mechanism for recalling several boards. Cognitive Psychology, 31(1), 1-40. (Extension of chunk theory: templates manage complex positions via flexible cognitive structures with fixed core and variable slots.)
  • Miller, G. A. (1956). - The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63(2), 81-97. (Working-memory capacity is universally limited to ~7 items; chunking is how expertise sidesteps the limit.)
  • Ericsson, K. A., & Chase, W. G. (1982). - Exceptional memory. American Scientist, 70(6), 607-615. (Domain-specific memory is built through deliberate practice, not innate talent.)
  • Charness, N. (1976). - Memory for chess positions: Resistance to interference. Journal of Experimental Psychology: Human Learning and Memory, 2(6), 641-653. (Adults build chunks as effectively as children at comparable learning rates.)