Working memory (WM) is the cognitive system that maintains and manipulates information in the service of ongoing cognition. Unlike the passive short-term store it replaced in theoretical models, working memory emphasizes the active processing, updating, and integration of information. It is central to virtually every complex cognitive ability — reasoning, language comprehension, mental arithmetic, problem solving, and learning — and individual differences in working memory capacity are among the strongest predictors of general intelligence.
Baddeley's Multi-Component Model
Alan Baddeley and Graham Hitch (1974) proposed the most influential model of working memory, replacing the unitary short-term store with a multi-component system. The central executive is an attentional control system that directs processing, coordinates the subsidiary systems, and manages the interface with long-term memory. The phonological loop holds and rehearses verbal and acoustic information through a phonological store (which holds memory traces for ~2 seconds) and an articulatory rehearsal process (inner speech). The visuospatial sketchpad maintains and manipulates visual and spatial information. The episodic buffer, added in 2000, integrates information from the subsidiary systems and long-term memory into coherent episodes.
Word length effect: fewer long words recalled than short words
Articulatory suppression: repeating "the" disrupts verbal WM
These three effects converge on a system with a sound-based store and an articulatory rehearsal mechanism.
Working Memory Capacity
Working memory capacity varies substantially across individuals and is strongly related to higher cognitive abilities. Complex span tasks (such as reading span, operation span, and symmetry span) measure WM capacity by requiring participants to maintain information while simultaneously processing other material. Performance on these tasks correlates strongly (r ≈ 0.5-0.7) with fluid intelligence, reading comprehension, and academic achievement.
The relationship between WM capacity and fluid intelligence is one of the most robust findings in individual differences research. Engle and colleagues have argued that the key component driving this correlation is executive attention — the ability to maintain task-relevant information in an active state in the face of distraction and interference. This view links WM capacity to the broader construct of cognitive control and suggests that WM limitations fundamentally constrain the complexity of thought.
Alternative Models
Other models of working memory include Cowan's embedded processes model, which treats WM as the activated portion of long-term memory plus a limited-capacity focus of attention (~4 items), and Oberauer's concentric model, which proposes three levels: activated LTM, a region of direct access (~4 items), and a single focus of attention (~1 item). These models emphasize the role of attention in WM and the close relationship between WM and long-term memory.
Neural Basis
Working memory engages a distributed network centered on the prefrontal cortex. The dorsolateral prefrontal cortex (DLPFC) is particularly important for maintenance and manipulation, while the ventrolateral prefrontal cortex is more involved in retrieval and selection. Sustained neural activity during delay periods — observed in both single-unit recordings and fMRI — has long been considered the neural signature of information maintenance in WM. Recent work has challenged this view, suggesting that WM may also rely on activity-silent mechanisms, such as changes in synaptic weights that can be reactivated by an appropriate probe.
WM Training
Whether WM capacity can be increased through training has been one of the most debated questions in cognitive psychology. Jaeggi and colleagues (2008) reported that training on an adaptive dual n-back task improved fluid intelligence, sparking enormous interest and commercial "brain training" programs. However, subsequent meta-analyses have generally found that while WM training improves performance on trained and similar tasks, transfer to fluid intelligence and other cognitive abilities is minimal or absent.