Evidence for How Memory Works: Working and Long-Term Memory #

Every substantive claim on the How Memory Works: Working and Long-Term Memory page is checked against current research. Here is each claim, how well today’s evidence supports it, and the sources. The full, de-duplicated source list lives on the references page.

Supported · moderate evidence — Memory can be usefully modelled as a flow from brief sensory memory, through a limited-capacity working memory, into a vast and durable long-term store.

The sensory/short-term/long-term staged model (Atkinson & Shiffrin) remains the standard textbook framework as of 2026; modern work refines rather than overturns it, treating the stages as a simplification of more interactive processes.

Sources: Atkinson, R. C. & Shiffrin, R. M. (1968), Human memory: A proposed system and its control processes, Psychology of Learning and Motivation · full reference ›

Supported · strong evidence — Working memory is a limited-capacity system that actively manipulates information rather than merely storing it passively, and is best described as multiple cooperating components.

Baddeley’s review of working-memory theory documents broad consensus that working memory is an active multi-component system (phonological loop, visuospatial sketchpad, central executive, episodic buffer); the component framing is mainstream in 2026 even as rival models exist.

Sources: Baddeley, A. (2012), Working memory: theories, models, and controversies, Annual Review of Psychology, 63, 1-29 · full reference ›

Supported · moderate evidence — The capacity of working memory for distinct chunks is roughly four items, not the older ‘seven plus or minus two’.

Cowan’s synthesis argues the pure capacity limit, once chunking and rehearsal are controlled, is about 3-5 items; this ~4-item estimate is widely cited in 2026, though the exact number remains debated and is task-dependent.

Sources: Cowan, N. (2010), The magical mystery four: how is working memory capacity limited, and why?, Current Directions in Psychological Science, 19(1), 51-57 · full reference ›

Supported · moderate evidence — Information not attended to or rehearsed is lost from working memory within seconds.

Rapid decay/displacement of unrehearsed short-term information is a long-standing, well-replicated finding (classic Brown-Peterson paradigm) and is consistent with capacity-limited accounts reviewed by Cowan and Baddeley.

Sources: Cowan, N. (2010), The magical mystery four, Current Directions in Psychological Science, 19(1), 51-57 · full reference ›

Supported · strong evidence — Remembering involves three distinct processes—encoding, storage and retrieval—and failures can occur at any one of them.

The encoding/storage/retrieval decomposition is foundational, uncontested cognitive-psychology framework as of 2026 and appears in every standard memory text.

Sources: Atkinson, R. C. & Shiffrin, R. M. (1968), Human memory: A proposed system and its control processes, Psychology of Learning and Motivation · full reference ›

Supported · strong evidence — Recently formed memories are initially labile and become more stable over time through consolidation, with sleep supporting this process.

Memory consolidation, and the role of sleep in stabilising newly encoded memories, is well established by 2026 (Rasch & Born review), though the precise mechanisms continue to be refined.

Sources: Rasch, B. & Born, J. (2013), About sleep’s role in memory, Physiological Reviews, 93(2), 681-766 · full reference ›

Supported · strong evidence — Actively retrieving a memory strengthens it and improves later retention more than restudying the same material (the testing effect).

The testing/retrieval-practice effect is one of the most robust and replicated findings in learning science as of 2026, supported by Roediger & Karpicke and numerous meta-analyses across materials and ages.

Sources: Roediger, H. L. & Karpicke, J. D. (2006), Test-enhanced learning: taking memory tests improves long-term retention, Psychological Science, 17(3), 249-255 · full reference ›

Supported · moderate evidence — Without review, retention of newly learned material drops rapidly at first and then more slowly—the forgetting curve.

Ebbinghaus’s forgetting curve has been directly replicated (e.g. Murre & Dros, 2015) and its broad shape—steep early loss, then flattening—is accepted in 2026; exact rates vary with material, meaningfulness and individual.

Sources: Murre, J. M. J. & Dros, J. (2015), Replication and analysis of Ebbinghaus’ forgetting curve, PLOS ONE, 10(7), e0120644 — https://doi.org/10.1371/journal.pone.0120644 · full reference ›

Supported · strong evidence — Re-reading produces a feeling of familiarity (fluency) that overstates actual learning, making it a weak study strategy relative to self-testing.

Dunlosky et al.’s large review rates rereading and highlighting as low-utility and practice testing as high-utility, and documents that fluency from rereading inflates confidence without commensurate retention; this is consensus in 2026.

Sources: Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J. & Willingham, D. T. (2013), Improving students’ learning with effective learning techniques, Psychological Science in the Public Interest, 14(1), 4-58 · full reference ›

Supported · strong evidence — Spacing retrievals over time (distributed practice) produces more durable learning than massing them together.

The spacing effect is supported by a large meta-analysis (Cepeda et al., 2006) and remains one of the best-evidenced learning principles in 2026, robust across domains and retention intervals.

Sources: Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T. & Rohrer, D. (2006), Distributed practice in verbal recall tasks: a review and quantitative synthesis, Psychological Bulletin, 132(3), 354-380 · full reference ›

Supported · moderate evidence — Breaking material into smaller chunks and reducing concurrent demands improves learning because working-memory capacity is limited.

That respecting working-memory limits (chunking, reducing load) aids learning follows from capacity research (Cowan) and is the basis of cognitive load theory, which is mainstream in 2026; effect sizes depend on task and prior knowledge.

Sources: Cowan, N. (2010), The magical mystery four, Current Directions in Psychological Science, 19(1), 51-57 · full reference ›