Knowledge base Updated 2 November 2026 1 views

Working Memory

The system that holds and manipulates a small amount of information over seconds, supporting reasoning, comprehension and learning. Capacity is limited, individual differences are large, and it predicts academic and cognitive outcomes well.

Definition

Working memory is distinguished from short-term memory by the requirement to do something with what is held, not merely retain it.

Baddeley and Hitch's multi-component model proposes:

a phonological loop for verbal material,
a visuospatial sketchpad for visual and spatial material,
a central executive allocating attention,
and, added in 2000, an episodic buffer integrating across codes and linking to long-term memory.

Competing accounts describe working memory as the activated portion of long-term memory plus attentional control, rather than as separate stores.

History

Miller's 1956 paper gave the field its most quoted number, seven plus or minus two. Miller himself framed it lightly, and the paper's argument about chunking is more important than the number.

Cowan's later analysis, controlling for rehearsal and chunking, put the pure capacity limit closer to four. The four estimate is the better-supported one, and the seven is the one everybody repeats.

Theory

The live disagreement is about what limits capacity.

Slot models: a fixed number of discrete item representations.
Resource models: a continuous resource divided among items, so precision falls as load rises.
Interference models: the limit arises from similarity-based interference and temporal decay rather than a store size.

Evidence exists for each and the argument has not resolved. What is agreed is that attentional control accounts for a large share of individual differences, and that working memory capacity correlates substantially with fluid intelligence — high enough that some researchers argue they are largely the same construct.

Research

Complex span tasks — operation span, reading span — predict reading comprehension, mathematics achievement and reasoning.

Working memory training does not generalise. This is one of the better-established negative findings in cognitive psychology. People improve at the trained task, improvements transfer to very similar tasks, and they do not transfer to intelligence, academic performance or everyday functioning. Meta-analyses using active control groups find near-zero far transfer, while the commercial brain-training industry continues on the opposite premise.

Applications

Instructional design, particularly the management of extraneous load; assessment in learning difficulties and ADHD; interface design where holding information across steps is the cost.

The practical implication of the training literature is to reduce the load a task imposes rather than to attempt to expand capacity.

Criticism

Seven plus or minus two is wrong and is still printed everywhere.
Model proliferation. Several models fit the data comparably, which usually means the data do not discriminate between them.
Task impurity. Complex span tasks recruit processing as well as storage, so a score is not a clean measure of a store.
Commercial overreach. Brain-training claims run far ahead of the evidence and have attracted regulatory action.

References

Baddeley, A. D., & Hitch, G. (1974). Working memory. Psychology of Learning and Motivation, 8, 47-89.
Miller, G. A. (1956). The magical number seven, plus or minus two. Psychological Review, 63(2), 81-97.
Cowan, N. (2001). The magical number 4 in short-term memory. Behavioral and Brain Sciences, 24(1), 87-114.
Melby-Lervåg, M., Redick, T. S., & Hulme, C. (2016). Working memory training does not improve performance on measures of intelligence. Perspectives on Psychological Science, 11(4), 512-534.

Written for PsychologyOrbit and reviewed before publishing. It is a summary for professionals and students, not clinical guidance, and it is not a substitute for the primary sources listed above.