Chunking
Grouping items into meaningful units to pack more into limited memory.
What it means
Chunking is the process of organizing individual elements into larger, meaningful units, so that working memory's narrow capacity holds more total information by storing fewer, richer chunks. Because the limit on short-term memory is roughly a fixed number of chunks rather than of raw items, recoding items into familiar patterns dramatically expands effective span. Expertise is built largely from vast libraries of domain chunks, which is why a chess master can reproduce a realistic board at a glance but not a random one. Chunking depends on long-term knowledge to define what counts as a unit, linking short-term performance to accumulated learning. It is a core mechanism behind mnemonic skill and skilled memory.
Seven, or maybe four
Miller's 1956 figure was offered half in jest — he described being persecuted by an integer — and it lumped together separate limits measured on different tasks. When later researchers controlled for subvocal rehearsal and for the grouping strategies that let people build larger units on the fly, the ceiling dropped. Cowan (2001) reviewed the convergent evidence and put the pure capacity at about four chunks, plus or minus one. The gap matters: a raw item span of seven usually reflects rehearsal and chunking already at work, not the underlying store. The honest summary is that working memory holds only a handful of chunks, and the exact count depends heavily on how tightly the material can be recoded.
What a chunk actually is
The word long stood for "whatever people treat as one unit," a definition that risks circularity. Two sharper accounts now compete. Mathy and Feldman (2012) recast a chunk as a unit in a maximally compressed code: material that can be described more briefly — runs, repeats, simple rules — is easier to hold, and span tracks compressibility rather than raw length. Norris and Kalm (2021) find the picture is mixed: for small two-word chunks the gain looks like redintegration — templates in long-term memory that help reconstruct a decaying short-term trace — while larger three-word chunks show genuine data compression that shrinks what must be stored. The two mechanisms predict different error patterns and remain actively debated.
It does not transfer
The most instructive limit comes from training studies. Ericsson, Chase and Faloon (1980) coached an undergraduate, SF, a keen runner, to recode digit strings as race times and ages; over roughly 230 hours of practice his digit span climbed from 7 to 79. But when they switched the material to random consonants, his span collapsed back to about six. The chunks he had built were made of running knowledge and applied only to numbers. Chase and Simon's chess masters show the mirror image: their memory edge vanishes on randomly arranged boards, where no familiar pattern fits. Chunking expands what you can hold only inside a domain you already know deeply; it never lifts the general ceiling.
Designing for it
The practical lever is to pre-package information into units the audience already reads as wholes. Account numbers, sort codes and software licence keys are hyphenated into threes and fours for exactly this reason, and multi-step instructions land better when related actions are named as one move rather than listed atomically. The caution is that "Miller's law" is routinely over-applied in interface design, cited to cap menus at seven items when the real constraint is how well the options group and label themselves, not their count. Because chunk size grows with expertise, the same layout that overwhelms a novice can feel spacious to an expert; design for the reader's existing chunks, not an abstract number.
Examples
The string FBICIAUSA is far easier to hold as the three chunks FBI-CIA-USA than as nine separate letters.
Phone numbers are printed in groups — 020 7946 0018 rather than eleven loose digits — because three chunks slip into working memory where eleven separate items will not.
An experienced cook reads a recipe once and holds it, because 'make a roux' is a single chunk to her and five separate instructions to a beginner.
A pianist sees a written C-major triad as a single hand shape, while a beginner reads three separate notes and hunts for each key in turn.
A memory athlete recodes each group of playing cards into one vivid character performing an action, so a shuffled deck becomes a short run of scenes rather than fifty-two loose cards.
First described in George Miller (1956); Chase & Simon (1973).
Key references
- Norris, D., & Kalm, K. (2021). Chunking and data compression in verbal short-term memory. Cognition, 208, 104534. doi.org/10.1016/j.cognition.2020.104534
- Mathy, F., & Feldman, J. (2012). What's magic about magic numbers? Chunking and data compression in short-term memory. Cognition, 122(3), 346-362. doi.org/10.1016/j.cognition.2011.11.003
- Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24(1), 87-114. doi.org/10.1017/S0140525X01003922
- Gobet, F., Lane, P. C. R., Croker, S., Cheng, P. C.-H., Jones, G., Oliver, I., & Pine, J. M. (2001). Chunking mechanisms in human learning. Trends in Cognitive Sciences, 5(6), 236-243. doi.org/10.1016/S1364-6613(00)01662-4
- Ericsson, K. A., Chase, W. G., & Faloon, S. (1980). Acquisition of a memory skill. Science, 208(4448), 1181-1182. doi.org/10.1126/science.7375930
- Chase, W. G., & Simon, H. A. (1973). Perception in chess. Cognitive Psychology, 4(1), 55-81. doi.org/10.1016/0010-0285(73)90004-2