Evidence for Cognitive load: working within your limits #

Every substantive claim on the Cognitive load: working within your limits 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 · strong evidence — The capacity of working memory for separate, unrelated chunks is limited to roughly three to four items, not the older ‘seven plus or minus two’ figure, once rehearsal and grouping are controlled for.

Cowan’s review and subsequent work converge on a pure working-memory capacity of about 3-4 chunks when chunking and rehearsal are prevented; this ~4 estimate remains the mainstream position in 2026, with the classic ‘7’ now understood as inflated by chunking strategies.

Sources: Cowan, N. (2010), The magical mystery four: How is working memory capacity limited, and why? Current Directions in Psychological Science — https://doi.org/10.1177/0963721409359277 · full reference ›

Supported · strong evidence — A ‘chunk’ of working memory can be a single element or a larger meaningful unit, so grouping items (e.g. letters into a word) reduces how many slots they occupy.

The chunk as the unit of working-memory capacity is a long-standing, well-replicated principle (Miller 1956; Cowan 2010); recoding several elements into one meaningful unit is a robust and uncontested finding.

Sources: Cowan, N. (2010), The magical mystery four — https://doi.org/10.1177/0963721409359277 · full reference ›

Supported · strong evidence — Demand on working memory during learning can be separated into intrinsic load (inherent element interactivity of the material), extraneous load (imposed by how material is presented), and germane load (effort that builds schemas).

The three-component model of cognitive load is the standard framework in instructional-design research and is laid out in Sweller, van Merrienboer & Paas’s 2019 review; it remains the dominant account in 2026, though the precise status and measurement of germane load continues to be refined.

Sources: Sweller, J., van Merrienboer, J. J. G., & Paas, F. (2019), Cognitive architecture and instructional design: 20 years later. Educational Psychology Review — https://doi.org/10.1007/s10648-019-09465-5 · full reference ›

Supported · moderate evidence — Reducing extraneous load (clutter, poor presentation, split sources of information) frees working-memory capacity for productive learning.

Effects such as the split-attention and coherence effects, where removing extraneous demand improves learning, are well supported by cognitive-load and multimedia-learning research, though effect sizes vary with learner expertise and task.

Sources: Sweller, J., van Merrienboer, J. J. G., & Paas, F. (2019), Cognitive architecture and instructional design: 20 years later — https://doi.org/10.1007/s10648-019-09465-5 · full reference ›

Supported · strong evidence — Schemas stored in long-term memory act as a single chunk when retrieved, so building schemas lets experts handle material that would overload a novice’s working memory.

Schema construction and automation as the mechanism by which expertise bypasses working-memory limits is a core, well-evidenced tenet of cognitive load theory and is consistent with classic expertise findings (e.g. chess-recall studies); it remains consensus in 2026.

Supported · moderate evidence — For novices, studying worked examples imposes less extraneous load and produces better learning than solving equivalent problems unaided (the worked-example effect).

The worked-example effect is one of the most replicated results in cognitive load theory, but it is moderated by expertise (the expertise-reversal effect): the advantage shrinks or reverses as learners gain schemas, which the page reflects by recommending self-solving once the pattern is familiar.

Supported · moderate evidence — Intrinsic load depends on how many interacting elements must be processed simultaneously, so sequencing learning from simpler parts to the whole keeps demand within working-memory limits.

That intrinsic load scales with element interactivity, and can be managed by isolating elements before combining them, is a standard and reasonably well-supported claim of cognitive load theory, though optimal sequencing is task-dependent.