You are viewing a free preview of this lesson.
Subscribe to unlock all 10 lessons in this course and every other course on LearningBro.
The Multi-Store Model (MSM), proposed by Atkinson and Shiffrin (1968), was the first comprehensive information-processing account of how human memory is structured. It represents memory as a flow of information through three separate stores — the sensory register, short-term memory (STM) and long-term memory (LTM) — linked by control processes such as attention, rehearsal and retrieval. Each store is characterised by its coding (the form in which information is held), its capacity (how much it can hold) and its duration (how long it can hold it), and these three properties, together with the studies that support and challenge them, form the examinable core of this topic. For Edexcel cognitive psychology the MSM matters twice over: it is a theory to be described and evaluated in its own right, and it is the structural benchmark against which the working memory model, Tulving's types of long-term memory and the explanations for forgetting are all understood.
Key Definition: The Multi-Store Model is a representation of memory as three unitary stores (sensory register, STM, LTM) connected by the processes that transfer information between them (attention, rehearsal, retrieval).
This lesson addresses the Edexcel 9PS0 — Paper 1, Topic 2: Cognitive Psychology content on the multi-store model of memory: the sensory register, short-term memory and long-term memory, and the coding, capacity and duration of each store, together with the processes that move information between them. It introduces the supporting and challenging research the specification expects for evaluation — Sperling, Miller, Jacobs, Peterson and Peterson, Baddeley (1966) and Bahrick et al. — and links forward to the working memory model and the types of long-term memory, both of which arose as critiques of the MSM. In assessment-objective terms, you should be able to describe the three stores and their features (AO1), apply the model to everyday and scenario-based examples of remembering (AO2), and evaluate it through controlled experiments, clinical case studies and competing models (AO3).
Connects to…
The diagram below sets out the full architecture of the MSM: the three stores, the processes that transfer information between them (attention, maintenance and elaborative rehearsal, retrieval), the rehearsal loop that keeps material active in STM, and the routes by which information is lost at each stage.
flowchart LR
INPUT["Environmental<br/>stimulus"] --> SR["Sensory Register<br/>(iconic, echoic)"]
SR -->|Attention| STM["Short-Term Memory<br/>STM"]
STM -->|Maintenance rehearsal<br/>(rehearsal loop)| STM
STM -->|Prolonged / elaborative rehearsal| LTM["Long-Term Memory<br/>LTM"]
LTM -->|Retrieval| STM
SR -.->|Not attended to:<br/>decay| LOST1["Forgotten"]
STM -.->|Not rehearsed:<br/>decay / displacement| LOST2["Forgotten"]
LTM -.->|Retrieval failure /<br/>interference| LOST3["Cannot be recalled"]
The sensory register is the first store; all incoming sensory information enters here automatically, without any conscious effort.
| Feature | Detail |
|---|---|
| Duration | Very brief — under half a second for visual (iconic) memory and roughly 2–4 seconds for auditory (echoic) memory |
| Capacity | Very large — it receives the full inflow of sensory input across all senses at once |
| Coding | Modality-specific — information is held in the raw sensory form in which it arrived (visual, acoustic, haptic, and so on) |
STM is the second store, holding a small amount of information temporarily for immediate use.
| Feature | Detail | Key Research |
|---|---|---|
| Capacity | Approximately 7 ± 2 items | Miller (1956), in "The Magical Number Seven, Plus or Minus Two", reviewed evidence that STM holds about five to nine items and showed that chunking raises the effective capacity by grouping items into meaningful units. |
| Capacity (digit span) | Around 7 letters; slightly more for digits | Jacobs (1887) used the serial digit-span technique and reported a mean span of about 9.3 digits and 7.3 letters — digits being easier because there are only ten of them (0–9). |
| Duration | About 18–30 seconds without rehearsal | Peterson and Peterson (1959) gave participants a consonant trigram (e.g. BKG) and had them count backwards in threes to prevent rehearsal. Recall was around 80% after 3 seconds but only about 3% after 18 seconds, showing STM fades rapidly without maintenance rehearsal. |
| Coding | Primarily acoustic (sound-based) | Baddeley (1966) found that immediate recall of acoustically similar words (cat, cab, can) was significantly worse than of acoustically dissimilar words, indicating STM codes acoustically. |
Key Definition: Maintenance rehearsal is the repetition of information to keep it active in STM. In the MSM, sufficient maintenance rehearsal is what transfers information into LTM.
LTM is the third store, capable of holding information for very long periods, potentially across a whole lifetime.
| Feature | Detail | Key Research |
|---|---|---|
| Capacity | Potentially unlimited | No study has established an upper bound to LTM capacity. |
| Duration | Up to a lifetime | Bahrick et al. (1975) tested recall of former high-school classmates. Using yearbook photographs, recognition remained around 90% even 15 years after graduation and around 70% after 48 years; free recall was poorer (around 60% at 15 years, 30% at 48 years) but still demonstrated extremely durable storage. |
| Coding | Primarily semantic (meaning-based) | Baddeley (1966) found that recall of semantically similar words (big, large, great) after a delay was significantly worse than of semantically dissimilar words, indicating LTM codes semantically. |
A defining feature of the MSM is that it treats memory as a linear and largely unidirectional flow: information must pass through each store in sequence — sensory register, then STM, then LTM — and the only major backward step is retrieval (bringing material from LTM into STM for conscious use). This sequential structure is precisely what makes the model so testable, because it generates clear predictions: information that is never attended to cannot reach STM; information that is never rehearsed cannot reach LTM; and damage to the transfer process between STM and LTM should leave a person able to hold information briefly yet unable to store anything new permanently — exactly the pattern seen in patient HM. The model's clarity is therefore also the source of its vulnerability, because each strong prediction becomes a point at which contrary evidence (such as the dual-task and levels-of-processing findings discussed below) can challenge it.
| Process | Description |
|---|---|
| Attention | Moves information from the sensory register into STM; without it, information decays |
| Maintenance rehearsal | Repetition that keeps information circulating in STM (the rehearsal loop) and, if prolonged, transfers it to LTM |
| Retrieval | Brings information back from LTM into STM for conscious use |
Key Definition: Chunking is the combining of individual items into larger, meaningful units to increase the effective capacity of STM.
Miller (1956) observed that although STM holds about seven items, the size of each item can vary. The string "F-B-I-C-I-A-B-B-C" is nine separate letters (beyond STM capacity) but can be recoded into three familiar chunks — "FBI", "CIA", "BBC" — which sit comfortably within capacity. Chunking draws on existing knowledge held in LTM to form meaningful groups, which is why experts recall far more domain-specific material than novices: a chess master encodes a board position as a few meaningful patterns rather than dozens of individual pieces. This carries a theoretical implication that complicates the MSM's neat linear flow: if chunking in STM depends on knowledge retrieved from LTM, then information must be travelling backwards from LTM to STM at the very moment it is first being encoded — which sits awkwardly with the model's largely one-directional architecture and hints at the more interactive relationship between the stores that later models would emphasise.
The MSM's real value for the exam lies partly in how cleanly it applies to everyday remembering, which is what an AO2 scenario stem tests. Consider a student revising a new set of French vocabulary. The words on the page first enter the sensory register as visual (iconic) input; if the student attends to them they pass into STM, where the student can hold only about seven at a time and where the trace will fade within roughly half a minute unless it is rehearsed. Reading the words aloud shows the acoustic coding of STM at work — the student "hears" the words in their head. To make the vocabulary durable, the student must rehearse it repeatedly (or, better, process it for meaning by linking each word to an image or an English cognate), transferring it into LTM, where it is coded semantically and can, in principle, last a lifetime. In the exam, forgetting a word mid-test would be modelled as retrieval failure from LTM back into STM. Being able to map a concrete scenario onto the model's stores and processes in this way — naming the coding, capacity and duration at each stage — is exactly the skill a scenario question rewards, and it demonstrates understanding beyond a memorised description.
A strength of the MSM is that Baddeley's (1966) coding research supports its claim that STM and LTM are qualitatively distinct stores. Recall errors on immediate (STM) tasks were dominated by acoustic confusions, whereas errors on delayed (LTM) tasks were dominated by semantic confusions — exactly the dissociation the MSM predicts if the two stores code information differently. This matters because it provides controlled experimental evidence that STM and LTM are not simply the same memory at different time points but genuinely different systems. The implication is that the model's most basic structural claim — separate short- and long-term stores — rests on solid empirical ground rather than intuition alone.
The model is further supported by case studies of patients whose brain damage selectively impairs one store. HM (Henry Molaison), after surgical removal of much of his hippocampus to treat severe epilepsy, retained a functioning STM and largely intact old long-term memories but could no longer transfer new information into LTM — a striking dissociation supporting the existence of separate STM and LTM stores with distinct neural bases. Clive Wearing, following viral encephalitis, shows the same profile: a few-seconds STM yet devastated long-term memory. This matters because such selective impairments are hard to explain unless STM and LTM are genuinely separate stores. The implication is that the model's architecture is corroborated by neuropsychological as well as experimental evidence — though, as noted below, case-study evidence carries its own caveats.
Further support for two separate stores comes from the serial position effect. When people are given a list of words to recall in any order, they typically remember the first few words well (the primacy effect) and the last few words well (the recency effect), but words in the middle poorly. The MSM explains this elegantly: the early words have been rehearsed and transferred into LTM (primacy), while the final words are still active in STM at the moment of recall (recency). Critically, Glanzer and Cunitz (1966) found that inserting a brief distractor task between presentation and recall — which should displace items still held in STM — abolished the recency effect but left the primacy effect intact, exactly as the model predicts if recency depends on a fragile short-term store and primacy on a durable long-term one. This matters because such a clean dissociation between the start and end of a list is difficult to explain unless two functionally distinct stores are operating. The implication is that the serial position curve, and especially the selective destruction of recency by a distractor, provides behavioural experimental evidence (complementing the coding and case-study evidence) for the MSM's central claim of separate STM and LTM.
A major limitation is that the MSM treats STM as a single, unitary store, whereas evidence shows it has separable components. Dual-task studies demonstrate that people can perform a visual task and a verbal task simultaneously with little interference, which is impossible if a single store handles both. This finding led Baddeley and Hitch (1974) to replace the MSM's STM with the multi-component working memory model (a phonological loop and a visuo-spatial sketchpad under a central executive). This matters because it shows the MSM is not merely incomplete but structurally wrong about STM. The implication is that, although the MSM correctly identifies a short-term store, its characterisation of that store as unitary has been superseded by a better-evidenced model.
Similarly, the MSM treats LTM as a single, undifferentiated store, but research indicates several distinct types. Tulving (1985) proposed that LTM comprises episodic (personal events), semantic (general knowledge) and procedural (skills) memory, each with different properties and brain locations — supported by patients such as Clive Wearing, whose procedural memory (playing the piano) survives even as episodic memory collapses. This matters because the MSM cannot explain how one "long-term store" can be selectively damaged in this way. The implication is that the MSM's LTM, like its STM, is an oversimplification that later, more differentiated accounts have had to correct.
The MSM claims that information enters LTM through prolonged maintenance rehearsal, but this is contradicted by evidence that depth of processing, not sheer amount of repetition, drives long-term retention. Craik and Lockhart (1972) argued in their levels-of-processing framework that information processed semantically (for meaning) is remembered far better than information processed shallowly, regardless of rehearsal, and Craik and Tulving (1975) found semantically processed words were recalled roughly three times better than structurally processed ones. Everyday experience agrees: emotionally significant or distinctive events are vividly remembered without any rehearsal at all. This matters because it challenges a central mechanism of the model. The implication is that the MSM's account of how information reaches LTM is, at best, incomplete — maintenance rehearsal is one route, but elaborative, meaningful processing is a more powerful one.
The evidence base has methodological weaknesses that temper the model's support. Much of the laboratory evidence (Peterson and Peterson; Baddeley's word lists) uses artificial stimuli — consonant trigrams and decontextualised word lists bear little resemblance to the meaningful information we remember in daily life — so the studies may lack ecological validity and tell us more about laboratory memory than real memory. Conversely, the case-study evidence (HM, Clive Wearing) involves unique individuals with idiosyncratic brain damage whose pre-injury memory function is usually unknown, so the findings may not generalise. The implication is that the MSM is supported by a mix of high-control-but-artificial experiments and rich-but-ungeneralisable case studies, and a cautious reading weighs both limitations rather than treating either source as decisive.
Weighing the evidence, the MSM was a landmark — the first testable, information-processing model of memory, which generated decades of productive research and provided the framework from which the working memory model and the types-of-LTM account grew. Its core claim of separate short- and long-term stores is well supported by coding research and neuropsychology. However, its treatment of both STM and LTM as unitary, and its over-reliance on maintenance rehearsal, are now known to be wrong, and better-evidenced successors have replaced it on these points. The defensible judgement is that the MSM is best regarded not as a correct model of memory but as a foundational and historically pivotal one whose very limitations directly motivated the more sophisticated models that followed.
Craik and Lockhart (1972) proposed the levels-of-processing (LoP) framework as an alternative emphasis. Rather than focusing on separate stores, LoP holds that retention depends on the depth at which information is processed during encoding.
| Level of Processing | Description | Example | Recall |
|---|---|---|---|
| Structural (shallow) | Processing physical appearance | "Is the word in capital letters?" | Poorest |
| Phonemic (intermediate) | Processing sound | "Does the word rhyme with 'train'?" | Moderate |
| Semantic (deep) | Processing meaning | "Does the word fit: 'The ___ was delicious'?" | Best |
Craik and Tulving (1975) confirmed experimentally that semantically processed words were recalled roughly three times better than structurally processed ones. LoP therefore reinforces a specific criticism of the MSM — that transfer to LTM depends on more than maintenance rehearsal — without itself constituting a complete model of memory, since it describes encoding but says nothing about where or how information is stored. It is also open to the charge that "depth" is circular: it is defined as processing that yields better recall, yet better recall is then explained by depth, with no independent measure of depth.
Specimen question modelled on the Edexcel 9PS0 paper format.
Evaluate the multi-store model of memory. (16 marks)
This 16-mark extended-response question is marked as roughly 6 marks AO1 (accurate, detailed description of the three stores, their coding, capacity and duration, and the processes linking them) and 10 marks AO3 (evaluation — support from coding research and case studies, the oversimplification of STM and LTM, the rehearsal critique, and the methodological limitations of the evidence). Application (AO2) marks would apply only if a scenario stem were provided. The top band requires evaluation that is sustained and integrated into a reasoned overall judgement rather than a list of isolated points.
The multi-store model was made by Atkinson and Shiffrin in 1968. It says memory has three stores. The sensory register takes in information from the senses and holds it for less than a second. If you pay attention it goes to short-term memory. STM can hold about 7 items (Miller) for 18–30 seconds (Peterson and Peterson) and codes acoustically. If you rehearse it, it goes to long-term memory, which is unlimited and lasts a lifetime and codes semantically.
There is evidence for the model. HM had brain damage and could not make new long-term memories but his STM was fine, which shows STM and LTM are separate. Baddeley found STM uses acoustic coding and LTM uses semantic coding, which also shows they are different stores.
One weakness is that the model is too simple because it says STM is one store, but the working memory model showed STM has different parts. Another weakness is that it says you need rehearsal to remember things, but Craik and Lockhart said deep processing is more important. Overall, the MSM was an important model but it is too simple.
Examiner-style commentary: To reach the next band this answer needs to explain why each criticism matters rather than just naming it — for example, that dual-task findings make the unitary STM not merely incomplete but structurally wrong. The description is accurate but thin: the coding of the sensory register and chunking are missing and Sperling is not cited. The AO3 is two brief points with no reasoned judgement, and given the 10:6 weighting towards evaluation this caps the mark.
The multi-store model (Atkinson & Shiffrin, 1968) describes memory as three stores linked by processes. The sensory register holds all incoming sensory information very briefly (iconic memory under a second, shown by Sperling, 1960); attention transfers it to STM. STM has a capacity of about 7±2 items (Miller, 1956), a duration of 18–30 seconds without rehearsal (Peterson & Peterson, 1959) and codes acoustically (Baddeley, 1966). Maintenance rehearsal keeps information in STM and, if prolonged, transfers it to LTM, which has potentially unlimited capacity, lifetime duration (Bahrick et al., 1975) and semantic coding.
The model has clear support. Baddeley's coding research shows STM and LTM use different codes, supporting the idea that they are separate stores. Case studies reinforce this: HM could form short-term but not new long-term memories, indicating distinct stores with different brain bases.
However, there are limitations. The MSM treats STM as one store, but dual-task studies show people can do a visual and a verbal task at once, which led to the working memory model (Baddeley & Hitch, 1974). It also treats LTM as one store, but Tulving showed there are episodic, semantic and procedural types. Finally, Craik and Lockhart (1972) argued that depth of processing, not rehearsal, determines memory, which challenges the MSM's transfer mechanism. Overall, the MSM is a useful foundational model but oversimplifies both STM and LTM.
Examiner-style commentary: The move into the top band is to stop listing strengths then weaknesses in parallel and instead build an integrated argument that separates the model's well-supported structural claim (two stores) from its oversimplified internal architecture. The description and studies are accurate and the three criticisms are relevant, but each is stated rather than developed through point–evidence–explanation, and the conclusion is a one-line summary rather than a reasoned judgement.
The multi-store model (Atkinson & Shiffrin, 1968) conceptualises memory as a linear flow through three structurally distinct stores. Sensory information enters the modality-specific sensory register, which has a very large capacity but a duration under a second for iconic memory, as Sperling's (1960) partial-report procedure demonstrated. Attention transfers information to short-term memory, which has a capacity of approximately 7±2 chunks (Miller, 1956; Jacobs, 1887), a duration of 18–30 seconds without rehearsal (Peterson & Peterson, 1959) and predominantly acoustic coding (Baddeley, 1966). Maintenance rehearsal sustains information in STM and, if prolonged, transfers it to long-term memory — a store of potentially unlimited capacity, lifetime duration (Bahrick et al., 1975) and predominantly semantic coding. The three stores are thus differentiated on coding, capacity and duration, and connected by the processes of attention, rehearsal and retrieval.
The model's foundational claim — that STM and LTM are separate stores — is well supported. Baddeley's coding research provides a clean experimental dissociation (acoustic confusions in STM, semantic confusions in LTM), and the neuropsychological cases of HM and Clive Wearing show selective impairment of new long-term memory with STM intact, which is hard to explain unless the stores are genuinely separate and neurally distinct. The serial position effect adds behavioural support: Glanzer and Cunitz (1966) found a distractor task selectively abolished the recency effect (STM) while sparing primacy (LTM). This convergence of experimental, behavioural and case-study evidence is a genuine strength.
Nonetheless, the model's internal architecture is now known to be oversimplified on two fronts. Treating STM as unitary is contradicted by dual-task evidence — performing a visual and a verbal task simultaneously without interference is impossible for a single store — which prompted the working memory model (Baddeley & Hitch, 1974); and treating LTM as unitary is contradicted by Tulving's (1985) episodic/semantic/procedural distinction, supported by Clive Wearing's preserved procedural memory amid devastated episodic memory. The MSM's transfer mechanism is also suspect: Craik and Lockhart (1972) showed that depth of processing, not sheer rehearsal, governs retention, with Craik and Tulving (1975) finding semantic processing roughly tripled recall over structural processing. These are challenges to central claims, not peripheral quibbles.
The supporting evidence must also be read critically. The controlled experiments (Peterson & Peterson; Baddeley) use artificial stimuli that lack the meaningfulness of everyday memory, limiting ecological validity; the case studies concern unique individuals with idiosyncratic damage and unknown pre-morbid functioning, limiting generalisability. The defensible overall judgement is therefore that the MSM was a pivotal and productive model whose core structural claim (separate short- and long-term stores) remains well evidenced, but whose unitary treatment of both stores and its rehearsal-based transfer mechanism have been corrected by better-supported successors. Its enduring importance lies less in being right than in being the testable framework whose weaknesses generated the working memory model and the multi-component account of LTM — a clear illustration of how a flawed model can drive scientific progress.
Examiner-style commentary: This answer is already in the top band; the only refinement would be to weigh the reductionism of the model explicitly as an evaluative frame. It is distinguished by sustained, integrated evaluation: it separates the model's well-supported structural claim from its oversimplified internal architecture, develops each critique through point–evidence–explanation–implication, scrutinises the methodology of the supporting evidence, and reaches a genuine judgement framing the MSM as pivotal-but-superseded. The discriminator throughout is the quality and connectedness of the AO3 reasoning rather than the quantity of description.
Contemporary memory research has moved well beyond the linear MSM while keeping its useful vocabulary. Neuroimaging has localised aspects of the model's stores — the hippocampus is now firmly implicated in consolidating new long-term memories (consistent with HM's deficit), while prefrontal regions support the manipulation of information in working memory — giving the originally abstract flow diagram a biological grounding that connects cognitive psychology to biological approaches. Modern accounts emphasise consolidation (the gradual stabilisation of memories over hours and during sleep) and reconsolidation (the finding that recalling a memory can briefly render it malleable), neither of which the static MSM anticipated, and both of which matter for understanding conditions such as PTSD and for the reliability of eyewitness memory studied in criminological psychology. A productive debate is whether memory is best conceived as a set of stores at all, or as a set of processes (encoding, consolidation, retrieval) operating on a single, distributed neural substrate — a question that pits the structural tradition of Atkinson and Shiffrin against more recent process-based and connectionist models, and that connects directly to the reductionism debate.
This content is aligned with the Edexcel A-Level Psychology (9PS0) specification.