Multi-Store Model

The Multi-Store Model of Memory

In 1968, Richard Atkinson and Richard Shiffrin proposed one of the most influential models in cognitive psychology: the multi-store model (MSM) of memory. The model proposes that memory is not a single system but consists of three distinct stores — the sensory register, short-term memory (STM), and long-term memory (LTM) — each with different properties in terms of capacity, duration, and encoding. Information flows between these stores in a fixed, linear sequence.

The Sensory Register

All information from the environment first enters the sensory register (SR), which holds raw, unprocessed sensory data briefly — for approximately 0.25 to 0.5 seconds. The SR has a very large capacity, holding everything detected by the senses at any given moment. Each sensory modality has its own store: the iconic store handles visual information and the echoic store handles auditory information. The vast majority of this information decays rapidly and is lost. Only information that is attended to passes on to STM.

Short-term Memory

Short-term memory (STM) has three defining characteristics that distinguish it sharply from the sensory register and from LTM. Its capacity is highly limited: George Miller (1956) famously demonstrated that STM can hold approximately 7 +/- 2 chunks of information — roughly 5 to 9 items, where chunking can extend functional capacity by grouping related items together. Its duration is brief — without active rehearsal, information is lost from STM within approximately 15 to 30 seconds (Peterson and Peterson, 1959). The dominant form of encoding in STM is acoustic — information tends to be represented in terms of sound, even when originally presented visually (Baddeley, 1966).

Information in STM can be maintained through maintenance rehearsal — mentally repeating information — but this primarily keeps information circulating in STM rather than guaranteeing long-term retention. Transfer to LTM is more reliably achieved through elaborative processing that engages meaning.

Long-term Memory

Long-term memory (LTM) is qualitatively different from STM in every key property. Its capacity is considered unlimited — there is no known upper bound on how much information can be stored. Its duration is potentially a lifetime, with memories retained for decades without rehearsal. The dominant form of encoding in LTM is semantic — information is stored primarily in terms of meaning — though LTM also encompasses episodic memories (personal experiences with temporal context) and procedural memories (how to perform skilled actions).

Evidence: The Serial Position Effect

Glanzer and Cunitz (1966) provided influential experimental evidence through the serial position effect. Participants recalled words from a list best at the beginning (primacy effect) and end (recency effect), but struggled with middle items. The primacy effect occurs because early items have had time to be rehearsed into LTM. The recency effect reflects items still present in STM at the time of recall. Crucially, when recall was delayed by a filler task, the recency effect disappeared whilst the primacy effect remained — demonstrating that the two effects reflect different memory stores, supporting the MSM's structural distinction.

Evidence: Patient HM

The case study of patient HM (Henry Molaison) provides compelling neuropsychological support. Following surgery that removed his hippocampus, HM could no longer form new long-term memories (anterograde amnesia), but his STM remained intact — he could hold a conversation but could not remember it minutes later. This double dissociation between preserved STM and impaired LTM formation strongly supports the MSM's proposal that these are functionally separate stores.

Evaluation

The MSM generated an enormous research programme and established the scientific study of memory as a structured enterprise. The STM-LTM distinction has strong empirical and neuropsychological support. However, the model has been criticised for oversimplifying the structure of STM — Baddeley and Hitch (1974) showed it is itself a complex, multi-component system. The model also overemphasises passive maintenance rehearsal as the route to LTM; Craik and Lockhart's (1972) levels of processing framework showed that depth of meaning-based processing is a far better predictor of long-term retention. Finally, treating LTM as a single entity overlooks Tulving's (1972) evidence for distinct episodic, semantic, and procedural memory systems.