Chunking Theory in Chess
by Bill Wall

Chess has one of the longest legacies of any game or sport and known for its intellectual depth, going back to 1905 and Alfred Binet in the first work on chess psychology. Chess has well-documented cognitive and educational benefits and how it can be used to study common abilities and tasks in life. The study of chess players can be related to mental imagery and problem-solving. By looking how strong chess players approach chess problems and positions in a game of chess, researchers can better understand cognitive processes like perception, memory, and visualization. One of these areas of research deals with the chunking theory in chess and how strong players show remarkable memory with their ability to reconstruct a chess position of many pieces after viewing it for just a few seconds.

Chunking in psychology is a process by which individual pieces of information are bound together into a meaningful whole. A chunk is defined as a familiar collection of more elementary units that have been inter-associated and stored in memory repeatedly and act as a coherent, integrated group when retrieved. For example, instead of remembering strings of letters such as E-L-O-F-I-D-E-U-S-C-F, chess players will find it easier to remember the chunks Elo-FIDE-USCF consisting of the same letters.

Chunking is a grouping together of a number of pieces and has been attributed as the main mechanism that differentiates novices from chess Grandmasters with regard to chess skill. Expert players acquire knowledge of chunks by studying master chess games. This knowledge assists the player to select the right move without the need to search many moves ahead.

A chunk in chess is a unit of information in long-term memory containing a meaningful grouping of some of the chess pieces on squares (POSs) that appear on a chessboard, plus associated moves and ideas. It is a patterned cluster of chess pieces. The action of chunking is the grouping together of chess pieces. Chess chunk examples would be a castled position (kingside or queenside), or a fianchettoed bishop, or a familiar pawn chain, or common bank-rank piece positions. A chunk representing a standard castled white king can be represented as the set {Rf1, Kg1, Pf2, Pg2, Ph2}. These five pieces can make up to 31 different chunks.

A castled king position with the knight developed in the f3 square can be represented as the set {Rf1, Kg1, Nf3, Pf2, Pg2, Ph2}, which represents one chess chunk. That is now 6 pieces that any moderate chess player can remember. Any good chess player can recognize all 6 pieces that can be put on 6 squares of the 64 possible squares on a chess board in seconds. But someone who is not familiar with the castled position with a knight develop to f3 would have trouble remembering it. I could even add 7 pieces with a fianchettoed bishop and a castled king with (Rf1, Kg1, Nf3, Pf2, Bg2, Pg3, Ph2} and any good chess player could recreate this position on a board in seconds after seeing the position in just a few seconds. This would represent another well-known chess chunk.

A chunk is a long-term memory symbol that has arbitrary subparts and properties, and can be used as a processing unit. Each chunk can be retrieved by a single act of recognition. According to chunking theory, first introduced by Chase & Simon in 1973, pairs of pieces that have numerous relations are more likely to be noticed together, or chunked. It takes at least two seconds to retrieve a new chunk of information. A chess master's vocabulary is over 50,000 acquired chunks, which is similar to the number of words used in many languages. A grandmaster has learned an estimated 300,000 chunks.

A chess player stores chunks in long-term memory, but the chunks are processed through short term or working memory. A master can store about 3-5 chunks of working memory at any one time. To visualize the entire board and hold in short-term memory requires a strong player seeing the entire chess position in only 3 chunks. Therefore, the chunks need to be quite large as more pieces are on the board.

It is proposed that chess masters perceive and remember 'chunks' of chess pieces in frequently occurring configurations. It is this knowledge that gives the master the ability to focus quickly on the right move when selecting a move. When confronted with a difficult chess position, a weaker player may calculate for half an hour, often looking many moves ahead, yet miss the right continuation. A grandmaster will see the right move immediately, without consciously analyzing anything at all. A novice will take more time and double check his moves, but a master will move faster (and have greater chunks in long-term memory), having confidence in his move decision.

Much of the evidence of chunking in chess is taken from psychological experiments such as Adriaan de Groot's memory test on master and novice chess players. Dozens of experiments have shown that chess masters (but not novices) are able to reconstruct meaningful chess positions after rapid glances almost perfectly, but cannot reconstruct with the same ease a random scrambling of the chess pieces found in a position. Masters only slightly do better than amateurs in reconstructing random chess positions after a few seconds of glancing at the position. Further experiments have shown that masters can reconstruct an unfamiliar game with 22 pieces after viewing it for 5 seconds with about 93% accuracy. Masters (rated over 2200) have an 81% accuracy. Experts (rated 2000 to 2200) have an accuracy of 72%. Class A players (1800-2000) have an accuracy of 51% and novices have an accuracy of 33%. It seems that real chess positions carry deep meaning, but random chess positions do not. Advanced chess players remember pieces in structured positions, and that pieces are remembered as groups or chunks rather than the individual pieces together.

Chess masters have access in their memory to a large database of these stored chess patterns, or chunks as they are called. These chess chunks are associated with plausible plans and ideas as they play over a game of chess and try to recall past patterns. A chunk usually consists of 3 to 6 pieces, and the size of the stored chunks is positively correlated with skill. Both novices and masters can retain as many as 7 (plus or minus 2) chunks in short-term memory, but the more skilled players can make better use of their short-term memory because they have the right chunks available, have more experience, and have practiced longer. Thus, the difference in chess skill is based on differences in the number and the size of the chunks stored in long-term memory.

Positions taken from master chess games (random chess positions could not be remembered very well) were examined by grandmasters and novices. Grandmasters were able to remember almost every pieces of a presented positions perfectly while novices could barely place 50% of the pieces correctly on a new board after examining the position between 2 to 15 seconds. Masters do not encode chess positions as isolated pieces. Rather, they see the positions as large, dynamic complexes which also include empty squares, because the empty places play an important role in reconstructing the exact position. The ability of the masters to encode these large complexes is based on the amount of knowledge and experience that they have acquired over time from lots of chess practices and study. Masters recall bigger chunks and more chunks than novices, and they store that information in their long-term memory.

From the original chunking theory, the concept of a chunk was later expanded to more complex and large retrieval structures, called templates. Templates are formed if positions re-occur frequently and can contain free variables such as pointers to symbols representing plans, moves, strategical and tactical concepts. The template theory is that masters encode knowledge as relations between chunks and store other information besides the location of the pieces. These common patterns of specific chess pieces on squares are fundamental for recognition.

Chess masters rely on chunks and templates and not on purely visual information of pieces on squares to retrieve relevant information from long-term memory. Furthermore, abstract-level relations (sacrificial combinations) and analogies from past positions and games become more important to the chess master. Chess masters and grandmasters excel because their representations are on a high level of abstraction and not based on chunking alone. Chess masters rely on abstract representations in their chess positions while non-experts concentrate on superficial abstracts.

To account why chess masters can play high quality chess games in blitz chess (2 to 5 minutes), masters access information in their long-term memory (LTM) rapidly by recognizing familiar patterns of chess pieces on the board. These common chess patterns act as cues that trigger access to the chess chunks. Because these chunks are associated with possible moves, masters can usually choose good or best moves with very little look-ahead search. Chess masters perform very well in recall tasks of possible chess positions that might show up in a real game because storing one chunk on short-term memory (STM) gives access to a number of pieces. However, it is too hard to chunk positions with no meaning or purely random. There is no pattern to base these random positions on. Recall of random positions is almost equal among masters and novices, with less than a 50% success rate.

Experiments have been done with unrated amateurs to strong masters on grouping positions (pair the chess positions which intuitively seemed similar) for 20 middlegame or endgame positions. The pieces and their respective positions on the chess board were very similar. Novices almost always paired positions based on visual similarity, while masters grouped the positions into abstract pairs, such as an endgame with bishops of the same color, endgames with bishops of opposite color, or material gain was possible due to a double attack. Masters were able to associate positions with future development and plans. Chess amateurs were unable to do this. What differentiated masters and novices was the level of abstraction at which positions can be represented.

Another experiment with a different group of experts and novices performed the same test as above, but was asked to label the characteristics of the position pairs that they paired. Most of the novices grouped their pairs as visual similarity. But the experts were able to group their pairs as checkmate in one move, endgame with opposite colored bishops, endgame with bishops of the same color, pawn endgame, passed pawn in the pawn ending. None of the novices were able to label and categorize the paired positions like the experts.

It is interesting to note that chess masters will describe a position by the type of opening or main strategic plans, but almost never mention clusters of pieces sharing relations of defense, attack, or proximity. Chunks of pieces are missing from their description, either because they are so self-evident to a master, or they may not have verbal labels for these perceptual units.

The chess master not only has more chunks in long-term memory, can retrieve the chunks faster, but have also stored many more move sequences. Chess masters have thousands of sequences of opening moves stored away in long-term memory. Grandmaster have thousands of opening variations, sometimes running 20 moves deep or more, committed to memory. There are also thousands of traps and winning combinations of moves that every master knows.

One aspect of chess is unexplained by chunking theory. Blindfold chess is where a player can play an entire chess game (sometimes multiple games) without looking at the chess board (and repeat all the moves after the game). Playing blindfold chess requires access in memory to a tremendous amount of information. Most novices are incapable of playing even a handful of moves blindfolded, never mind a whole game. However, many masters and grandmasters can do it effortlessly. There have been several players who have played over 40 chess boards simultaneously blindfolded. In fact, studies have shown that there are no significant differences between chess play during rapid (15 minutes) blindfold chess and regular rapid chess among grandmasters. According to chunking theory, the master's short-term memory would be completely filled at any given point in a blindfold game, so the master should not be able to calculate or plan, but this is not the case.

The key to becoming a strong player is to learn chess positions, and learn lots of them. Learn what moves work in these positions and what moves are bad or contribute to anything in these positions. The learning of patterns or chunks is dependent on repetition and timing. The patterns to learn are the opening patterns, early middlegame patterns, positional patterns, tactical patterns, strategic endgame patterns, and technical endgame patterns. My books on chess miniatures and traps would be a good start to look at opening patterns and tactical patterns.


Adesman and Frey, Recall Memory for Visually Presented Chess Positions, Memory & Cognition, 1976
Atlas, Berger, Cooke and Lane, The role of high-level knowledge in memory for chess positions, American Journal of Psychology, pp. 321-351, 1993
Berger, Cognitive Organization in Chess: Beyond Chunking, AFIT Thesis, 1989
Bilalic and Gobet, The influence of instruction on chess expert perception, Cognitive Science, 33(5) pp. 743-747, 2009
Brum and Linhares, Understanding Our Understanding of Strategic Scenarios: What Role Do Chunks Play? Cognitive Science, 31, pp. 989-1007, 2007
Chabris and Hearst, Visualization, pattern recognition, and forward search: Effects of playing speed and sight of the position on grandmaster chess errors, Cognitive Science, 27, 637,648, 2003
Chase and Simon, Perception in chess, Cognitive Psychology, 4(1), pp. 55-81, 1973
Cook, Chunk Learning and Move Prompting: Making Moves in Chess, 2008
Cook, Computational Chunking in Chess, PhD Thesis, 2011
de Groot, Thought and Choice in Chess, 1946, 1965, and 1978
Freyhoff, Gruber, and Ziegler, Expertise and hierarchical knowledge representation in chess, International Symposium on the Psychology of Chess Skill, 1990
Gilmartin and Simon, A Simulation for Chess Positions, Cognitive Psychology, vol 5, pp. 29-46, 1973
Gobet and Simon, Templates in Chess Memory, Cognitive Psychology, 31(1), pp. 1-40, 1996
Gobet and Simon, Expert Chess Memory: Revisiting the chunking hypothesis, Memory, 6(3), pp. 225-255, 1998
Gobet, Lane, and Smith, Checking Chess Checks with Chunks: A Model of Simple Check Detection, Proceedings of the 9th International Conference on Cognitive Modeling, 2009
Gold, Opwis, and Schneider, Chess expertise and memory for chess positions in children and adults, Journal of Experimental Child Psychology, 1992
Gordiievych, Human-like Chess Playing Program, 2015
Holding, Theories of chess skill, Psychological Research, 54, pp. 1016, 1992
Jake and Leon-Villagra, Categorization and Abstract Similarity in Chess
Linhares and Brum, What role do chunks play? Cognitive Science, 31(6), pp. 989-1007, 2007
Rohrer and Wixted, Analyzing the dynamics of free recall, Psychonomic Bulletin & Review, 1, pp. 89-106, 1994
Roring, Reviewing Expert Chess Performance: A Production-Based Theory of Chess Skill, PhD Dissertation, 2008
Saariluoma, Location coding in chess, Quarterly Journal of Experimental Psychology, 47A, pp. 607-630, 1994
Simons, How experts recall chess positions, 2012

Return to Main Page

Please report broken or duplicate links to the Webmaster.

Official Website
Copyright 2015 by William D. Wall
All Rights Reserved

Bill Wall