4.1.2 - The multi-store model of memory

What the model claims

The multi-store model of memory, often linked with Atkinson and Shiffrin (1968), is a structural model. It explains memory as a sequence of separate stores, with information moving between stores through attention, rehearsal and retrieval.

For each store, AQA expects three features. Coding means the form in which information is held; capacity means how much information can be held; duration means how long information can remain there.

The route is simple:

• Environmental information first enters the sensory register.
• If attention is paid, selected information moves into short-term memory (STM).
• If information is rehearsed, it is more likely to be transferred to long-term memory (LTM).
• When stored knowledge is needed, it is retrieved from LTM back into STM so it can be consciously used.

[DIAGRAM: asset_name: Multi-store model pathway; asset_slug: 4_1_2_01_the_multi_store_model_of_memory__diagram_01; description: The sensory register, STM and LTM, with attention, rehearsal and retrieval arrows and qualified coding, capacity and duration estimates.]

For AQA, the key AO1 is not just naming the three stores. It is being able to describe the features of each store: coding, capacity and duration. AO2 usually asks you to apply the route to a person trying to notice, hold or retrieve information. AO3 asks whether the model is useful, supported by evidence, or too simple.

The sensory register

The sensory register is the brief first store for information arriving from the senses. It is not one single store in the everyday sense. It is better understood as a set of modality-specific registers: visual information is held in iconic memory, auditory information in echoic memory, and so on.

The sensory register has:

• Coding: modality-specific. Information is held in the form in which it arrived, such as visual, auditory, tactile or olfactory information.
• Capacity: very large. Far more information enters the sensory register than we can consciously process.
• Duration: extremely brief. Visual sensory memory is often measured in fractions of a second, while auditory sensory traces may last a little longer.

Sperling's (1960) partial-report research is useful evidence for the sensory register. Participants briefly saw grids of letters. In whole-report trials they could usually report only a few letters, but in partial-report trials a tone told them which row to report just after the display disappeared. Participants could report most letters from whichever row was cued. Because the row was selected only after the display vanished, this implied that much of the whole array had briefly been available. It does not mean that reporting one row produced more letters than reporting the whole array. Delaying the cue reduced the partial-report advantage, showing rapid loss of the visual trace.

For example, if Maya glances at a packed noticeboard, lots of visual information reaches her sensory register. Only the bit she attends to, such as the exam-room number, is likely to move into STM. The rest disappears very quickly.

Short-term memory

Short-term memory is the limited-capacity store for information currently in conscious awareness. In the multi-store model, STM receives selected information from the sensory register and also receives retrieved information from LTM.

The core STM features are:

• Coding: mainly acoustic. Baddeley (1966) found that immediate recall was disrupted more by acoustically similar words than by semantically similar words, supporting the idea that STM often codes verbal material by sound.
• Capacity: limited. Miller (1956) famously estimated around seven plus or minus two chunks, and Jacobs' earlier digit-span work showed immediate span is limited and varies with material and age.
• Duration: brief without rehearsal. Peterson and Peterson (1959) used consonant trigrams and a counting-backwards task to prevent rehearsal; recall fell sharply across delays up to 18 seconds. This supports a brief STM duration without rehearsal, conventionally estimated at about 18–30 seconds. The counting task and repeated trigrams can also produce interference, so the result does not isolate passive decay alone.

The word "chunk" matters. A chunk is a meaningful unit. The digits 2, 0, 2 and 7 may be four separate items, but "2027" can become one chunk if a student links it to the first A-level exam year for this specification.

For example, if Aisha silently repeats a six-digit code while walking to a locker, she is using maintenance rehearsal to keep the code in STM. If she is distracted before entering it, the code may disappear because STM is brief and vulnerable.

Long-term memory

Long-term memory is the store for information that has been retained beyond the immediate present. In the multi-store model, information gets into LTM mainly through rehearsal in STM, and it is brought back into STM when we need to use it.

The core LTM features are:

• Coding: mainly semantic. Meaning is especially important for long-term retention. Baddeley's work on word similarity is often used to support the contrast between acoustic STM and more semantic LTM.
• Capacity: potentially unlimited. AQA students should avoid claiming that a precise upper limit has been found. The safer phrasing is that no practical capacity limit has been established.
• Duration: potentially lifelong. Bahrick, Bahrick and Wittlinger (1975) tested memory for former classmates using yearbook photos and names across retention intervals of many years, showing that some long-term information can remain accessible for decades.

LTM is not the same as perfect recall. A person may have stored information but fail to retrieve it because the right cue is missing, or because later learning interferes. Those explanations belong in a later lesson, but they help prevent a common mistake: "long-term" does not mean "always easy to remember."

Bahrick's evidence comes from meaningful, well-learned material. Later contact with classmates or looking at photographs may refresh these memories, so it supports long duration without showing that every unrehearsed memory lasts equally long.

For example, when Tom is asked to define "coding" in class, the definition is retrieved from LTM into STM. He can then hold it consciously long enough to answer.

Explaining serial-position effects

The pattern of recall across a word list can reveal which memory store is contributing to performance.

A serial-position effect is a relationship between an item's place in a list and its likelihood of recall. Better recall of early items is the primacy effect; better recall of the last items is the recency effect. The model explains these using different stores.

Early items receive more rehearsal before further words compete for attention, making transfer to LTM more likely. Recent items can still be in STM when recall starts. Middle items receive less rehearsal than early items and are more likely to be displaced by later items in the limited-capacity STM store.

Compare immediate recall with recall after a distracting task lasting longer than STM's usual duration. Immediate recall may show both primacy and recency. During the delay, the distractor prevents rehearsal, so the last words are lost from STM and recency falls. Early items already transferred to LTM can remain available, so primacy is relatively preserved. Explain both conditions; the names of the effects alone do not explain them.

For example, suppose immediate recall is best for the first and last few words, while recall after counting backwards for 45 seconds is best only for the first few. The selective loss of the last words supports stores with different durations. It does not prove that there are exactly three stores. To compare the conditions fairly, keep list length, presentation rate and recall time the same; otherwise a harder list or shorter opportunity to answer could explain the difference.

This also explains ordinary memory: rehearsing a password keeps it active in STM, repeated processing may support LTM, and using it later requires retrieval back into STM.

Evaluation and usefulness

The multi-store model is useful because it gives a clear, testable account of memory. It separates stores by coding, capacity and duration, which helps explain everyday differences between briefly holding a phone number and remembering a childhood address.

One strength is evidence for separate STM and LTM. Case studies such as Scoville and Milner's (1957) report of HM suggest that forming new conscious long-term memories can be severely impaired while immediate memory is relatively preserved. HM did not lose every kind of learning, so this is evidence for a distinction, not for one uniformly damaged LTM box. Shallice and Warrington's (1970) patient KF showed the opposite pattern in verbal memory: very poor short-term verbal span with more preserved long-term learning. Double dissociations like this support the idea that STM and LTM are not the same store.

A second strength is that the model generated research. Studies on coding, capacity and duration became easier to organise because researchers could ask which store was being tested. This is why the model remains a useful starting point in AQA memory, even though later models are more detailed.

The main limitation is that the model is too simple. It treats STM and LTM as unitary stores, but later research suggests STM has different components and LTM includes different kinds of memory. This matters because a single-box account cannot fully explain why someone might have poor verbal STM but stronger visual-spatial memory, or remember facts differently from skills.

Another limitation is the model's emphasis on rehearsal. Craik and Lockhart's levels-of-processing approach challenged the idea that repetition is the key route to durable LTM. Deep, meaningful processing often helps long-term recall more than simply repeating information. For revision, this is the difference between repeating "semantic coding" ten times and linking it to the idea of meaning, examples and exam scenarios.

There is also a methodological caution. Much supporting research uses artificial tasks such as recalling trigrams, digits or word lists. These tasks give good control, which is useful for scientific testing, but they may not fully represent richer everyday memory.

Taken together, the evidence suggests that the multi-store model is a strong introductory model: it explains basic differences between stores and has research support, but it is reductionist because it compresses complex memory processes into a simple linear pathway.

Quick recap

The sensory register holds brief modality-specific traces; attention selects information for STM. STM is limited and brief without rehearsal, while LTM has a much longer duration and no established practical capacity limit. Rehearsal and retrieval link the stores. Serial-position and patient evidence support distinctions, while specialised components and meaningful processing challenge the model’s simplicity.