Working Memory Capacity¶

Status: established
Last updated: 2026-06-01
Sources: Magicnumberseven Miller1956.Pdf, S0140525X01003922.Pdf
Tags: [working-memory, short-term-memory, memory-capacity, chunking, focus-of-attention, attention, cognitive-foundations]

Summary¶

Working memory capacity is the limit on how much information a person can hold in an immediately accessible state. Miller (1956) proposed that the span of immediate memory is roughly seven items, "the magical number seven, plus or minus two", and argued that this limit is set by the number of items (chunks) rather than the amount of information they carry, so recoding small units into larger chunks raises effective capacity. Cowan (2001) reconsidered the evidence and argued that the real, "pure" capacity limit is smaller — about four chunks — once grouping, rehearsal, and long-term memory are prevented from inflating the count. The two estimates are not contradictory so much as measurements of different things: a compound limit that allows chunking (about seven) versus a pure limit on the focus of attention (about four). The construct underlies the limited-capacity premise running through Information Processing, Mental Workload, and Situation Awareness.

Body¶

Context¶

This article rests on two foundational papers in the cognitive psychology of memory. Miller (1956), in The Magical Number Seven, Plus or Minus Two, examines limits on the capacity to process information, drawing together absolute-judgment experiments and immediate-memory experiments under the then-new framework of information theory. Cowan (2001), in a Behavioral and Brain Sciences target article, reconsiders that evidence four decades later and argues for a lower, more precise capacity limit. Within this knowledge base the article supplies the capacity premise that the cognitive-ergonomics core depends on: it grounds the working-memory stage described in Information Processing, the limited-resource account of Mental Workload, and the working-memory constraint on Situation Awareness, and it connects to the absolute-judgment / signal-detection material in Psychophysics And Signal Detection Theory.

Key Points¶

Miller (1956) first establishes a limit on absolute judgment. Across unidimensional absolute-judgment tasks — judging the pitch, loudness, or position of a stimulus — observers transmit a roughly constant amount of information, with a channel capacity of about 2.5 bits, corresponding to roughly six or seven distinguishable categories (PDF pp. 1–8, orig. pp. 81–90). Miller treats this span of absolute judgment as a real limit measured in bits of information, and notes a related discontinuity in the "span of attention": in number-estimation experiments observers subitize accurately up to about six dots before shifting to estimation (PDF p. 8, orig. p. 90).

Miller (1956) then separates this from the span of immediate memory, and this distinction is the paper's central contribution. The span of immediate memory is about seven items, but unlike absolute judgment it is limited by the number of items, not the amount of information: across test materials from binary digits to words, the span stays near seven even though information per item varies widely, so the amount of information recalled rises with information per item rather than staying constant (PDF pp. 9–10, orig. pp. 91–93). Miller captures the distinction with the terms bit and chunk — the number of bits is constant for absolute judgment, the number of chunks is roughly constant for immediate memory. Because the span is a fixed number of chunks, recoding the input into fewer, larger chunks (grouping binary digits into decimal, dits and dahs into letters and words) increases the information held within the same span; he calls this recoding and treats it as a general and significant device for extending capacity (PDF pp. 11–13, orig. pp. 93–96). Miller is explicit that the recurrence of seven in both spans is partly coincidence, and that chunking, not the specific number, is the deeper point (PDF pp. 9, 15, orig. pp. 91, 97).

Cowan (2001) argues that the true capacity limit is smaller than seven. He distinguishes pure capacity-based estimates from compound estimates: the classic seven-item digit span is a compound limit because it lets participants form larger chunks through rehearsal, grouping, and long-term memory, so the number of independent chunks is unknown (PDF pp. 1–2, orig. pp. 87–88). A pure capacity limit can be observed only under conditions that fix what the chunks are and block recoding — Cowan sets out four: (1) information overload that limits chunks to individual stimulus items, (2) steps that specifically block recoding into larger chunks, (3) performance discontinuities produced by the capacity limit, and (4) indirect effects of the limit (PDF pp. 2–3, orig. pp. 88–89). Across procedures meeting these conditions, the evidence converges on a mean capacity of about three to five chunks — roughly four on average — with individual scores ranging from about two to six (PDF p. 28, orig. p. 114). Cowan attributes this limit to a capacity-limited focus of attention rather than to a dedicated short-term store, and treats other, non-capacity-limited sources (such as sensory memory and rehearsal) as what inflate compound estimates.

Conclusion¶

Miller (1956) concludes that immediate memory holds a fixed number of chunks rather than a fixed amount of information, so the effective capacity depends on how richly the input is recoded; the famous seven is a convenient figure rather than a hard constant. Cowan (2001) concludes that once chunking and rehearsal are controlled, a single central limit of about four chunks emerges, identified with the focus of attention, while the higher seven-item figure remains valid as a commonly observed compound limit for materials that permit grouping. The two papers therefore agree on the core mechanism — capacity is counted in chunks, and chunking is what lets observers exceed the raw limit — and differ on the size of the pure limit because they isolate it differently.

Conflicts¶

Miller (1956) and Cowan (2001) give different numbers for the capacity of immediate memory: about seven items versus about four chunks. Cowan argues the discrepancy is methodological rather than substantive — the seven-item span is a compound estimate that allows undetected chunking, whereas the four-chunk limit is the pure limit observed when grouping and rehearsal are blocked (PDF pp. 1–2, orig. pp. 87–88). On this reading the lower figure is the more fundamental capacity limit and the higher figure is a real but inflated performance ceiling. Whether a single fixed number captures the limit at all remains contested in the peer commentary published with Cowan's article.

References¶

Cowan, N. (2001) 'The magical number 4 in short-term memory: A reconsideration of mental storage capacity', Behavioral and Brain Sciences, 24(1), pp. 87–114. doi: 10.1017/S0140525X01003922. cowan2001magical

Miller, G.A. (1956) 'The magical number seven, plus or minus two: Some limits on our capacity for processing information', Psychological Review, 63(2), pp. 81–97. doi: 10.1037/h0043158. miller1956magical

Open Questions¶

Is working memory capacity best expressed as a fixed number of chunks (Cowan, 2001) or as a flexible resource shared among items, and how should the two views be reconciled?
How does chunk size interact with chunk number under the four-chunk limit — does richer recoding raise effective capacity without raising the chunk count, as Miller (1956) implies?
How do capacity limits established with simple verbal materials transfer to the dynamic, multi-element displays of applied monitoring tasks (see Situation Awareness, Mental Workload)?