Cognitive Psychology
Working Memory
Baddeley's multi-component model — and the four-system architecture behind cognition
Working memory is the limited-capacity system that temporarily holds and manipulates information for ongoing cognitive tasks. Alan Baddeley and Graham Hitch's 1974 paper replaced the unitary short-term memory with a multi-component model: a central executive controlling two slave systems — the phonological loop for verbal material and the visuospatial sketchpad for visual-spatial material. Baddeley added the episodic buffer in 2000 to handle multi-modal binding. Capacity is roughly four chunks (Cowan 2001) rather than the seven Miller proposed in 1956. The system underwrites reading, mental arithmetic, problem-solving, and language comprehension, and its deficits feature in ADHD, dyslexia, and aging.
- Original modelBaddeley & Hitch (1974)
- Four componentsCentral executive, phonological loop, visuospatial sketchpad, episodic buffer
- Episodic buffer addedBaddeley (2000)
- Capacity~4 chunks (Cowan 2001)
- Earlier estimateMiller's 7±2 (1956)
- Critical regionsDorsolateral prefrontal cortex, parietal
Interactive visualization
Press play, or step through manually. The visualization is yours to drive — try it before reading on.
Watch the 60-second explainer
A condensed visual walkthrough — narrated, captioned, under a minute.
Why working memory matters
- Reading comprehension. Holding context across sentences requires the phonological loop.
- Mental arithmetic. Multi-digit calculation taps both loop and executive.
- Problem-solving. Complex span correlates 0.5-0.7 with fluid intelligence.
- Education. Capacity predicts academic achievement across age groups.
- Aging. Working memory declines reliably from age 30 onward.
- ADHD and dyslexia. Working memory deficits feature in both diagnoses.
- Interface design. Limit menu items and dialog complexity to four chunks.
Common misconceptions
- Capacity is 7±2. Cowan's revised four-chunk estimate is closer to true capacity.
- It's the same as short-term memory. Working memory adds active manipulation and multiple stores.
- Training raises general intelligence. Far-transfer benefits are weak or absent.
- The phonological loop holds meaning. It holds sound; meaning lives elsewhere.
- The model is finalized. The episodic buffer remains theoretically contested.
- Capacity is uniform across people. Individual differences are large and stable.
Frequently asked questions
What's the phonological loop?
A two-part subsystem for verbal material. The phonological store holds speech-based information for one to two seconds. The articulatory rehearsal mechanism refreshes the trace by silent inner speech, extending its lifespan. Baddeley's evidence comes from word-length effects (longer words rehearse more slowly, reducing span) and articulatory suppression (saying "the the the" prevents rehearsal and impairs verbal span). Damage to the loop produces specific verbal short-term memory deficits.
What's the visuospatial sketchpad?
The visual analog of the phonological loop, holding visual and spatial information. Logie (1995) further subdivided it into a visual cache (objects, color, shape) and an inner scribe (spatial location and movement). Tasks tapping it include mental rotation, navigation, and remembering arrangement. Damage produces visual-spatial impairment with intact verbal short-term memory — a double dissociation supporting separate systems.
What does the central executive do?
The most contested component. Baddeley conceived it as an attention controller that coordinates the slave systems, switches tasks, updates contents, and inhibits irrelevant material. Miyake et al. (2000) factor-analyzed executive tasks and identified three sub-functions: shifting, updating, and inhibition. Modern accounts often replace "central executive" with these distinct executive processes plus working memory storage.
Why was the episodic buffer added?
The original three-component model could not explain how phonological and visual information bind into integrated experiences — remembering a sentence as both sound and meaning, or a face with a name. Baddeley (2000) added the episodic buffer as a multi-modal integration store linked to long-term memory. The buffer is more controversial than the slave systems, but accounts for chunking, semantic enrichment, and conscious experience.
How does it differ from short-term memory?
Short-term memory in the older Atkinson-Shiffrin (1968) model was a unitary passive store. Working memory adds active manipulation, multiple specialized stores, and explicit attention control. The N-back task, complex span tasks (operation span, reading span), and dual-task paradigms measure working memory specifically. Simple digit span measures only the phonological loop component, not full working memory.
What's the actual capacity?
Miller's 7±2 (1956) measured digit span — really phonological loop capacity. Cowan (2001) reanalyzed across paradigms and proposed a true working memory capacity of about four independent chunks. Individual differences are substantial — capacity correlates around 0.5-0.7 with fluid intelligence (Engle, Kane, Tuholski 1999). Capacity is one of the strongest cognitive predictors of academic achievement.
Does working memory training transfer?
Mostly no. Direct training on N-back or complex span tasks improves performance on those tasks but rarely transfers to fluid intelligence, reading, or arithmetic. Melby-Lervåg, Redick, and Hulme's (2016) meta-analysis showed near-transfer but no reliable far-transfer. Programs like Cogmed produce task-specific gains, not general cognitive enhancement. Working memory capacity appears highly stable across the lifespan.