HookThe story that got shorter, tidier and more English every time
In 1932 the Cambridge psychologist Frederic Bartlett handed his students a Native American folk tale called The War of the Ghosts — a story stuffed with unfamiliar names, abrupt jumps and a supernatural ending that made no sense to an English undergraduate. He asked them to read it twice, then recall it, again and again, over hours, days and months. Almost nobody reproduced it accurately. Canoes drifted into 'boats'. 'Something black came out of his mouth' hardened into 'he foamed at the mouth'. The ghosts quietly disappeared. Every retelling was shorter, tidier and more English than the one before. Bartlett had caught the mind doing something a tape recorder can never do: quietly rewriting the past so that it makes sense.
That single finding is the spine of the AQA Memory topic — the idea that memory is not a recording you play back but a process you rebuild every time you remember. The specification walks you through it in three linked stages: the processes every memory passes through (encoding, storage, retrieval), the structures those processes run on (the multi-store model, with its sensory register, short-term store and long-term store), and the evidence that retrieval is an active reconstruction rather than a readout. Get all three and you can explain why you can sing a chorus you have not thought about in a decade, yet lose a stranger's name ten seconds after hearing it — and, more usefully for the exam, why two honest witnesses to the same event can swear to different versions of it.
ModelThe three jobs every memory has to do
Before any theory, learn the three processes, because every exam question about forgetting is secretly asking which one failed. Encoding is the input stage — turning what your senses take in into a form the brain can hold. That form has flavours: acoustic encoding stores things by their sound (which is why you rehearse a phone number by muttering it), visual encoding by their appearance, and semantic encoding by their meaning. Storage is holding the encoded information over time, whether that is seconds or a lifetime. Retrieval is the output stage — getting it back out, either by recall (producing it from nothing, like an essay) or by recognition (picking it out, like a multiple-choice question, which is why recognition feels easier).
The reason this matters is that 'I forgot' is ambiguous. Did the information never encode properly because you were not paying attention? Did it decay in storage? Or is it stored fine but you cannot find the retrieval cue — the tip-of-the-tongue feeling where recognition would instantly succeed? A strong answer names the stage. Naming the stage is the difference between describing forgetting and explaining it.
ModelThe multi-store model — three boxes and the arrows between them
The multi-store model (proposed by Atkinson and Shiffrin) says memory is not one thing but three separate stores, connected in a fixed line. Information hits the sensory register first: an enormous-capacity but near-instant store, one channel per sense, holding a raw copy of what you just saw or heard for a fraction of a second. Pay attention to any of it and that fragment passes into short-term memory (STM). STM is tiny and fragile — a capacity of about 7 items (give or take two) and a duration of roughly 18 to 30 seconds unless you keep it alive by rehearsal. Rehearse it enough and it transfers into long-term memory (LTM): effectively unlimited in capacity and lasting anywhere from minutes to a lifetime. Retrieval is the arrow running backwards — pulling information from LTM into STM so you can use it.
The three stores differ on three measurable features, and examiners love the grid. Coding: the sensory register holds raw sensory traces, STM codes mainly acoustically (we mix up letters that sound alike, like B and P), and LTM codes mainly semantically (we mix up words that mean similar things). Capacity: sensory register huge, STM about 7 items, LTM unlimited. Duration: sensory register under a second, STM under about 30 seconds, LTM up to a lifetime. Learn those three rows and you can describe any store on demand.
DataMurdock's serial position curve — the graph that proves there are two stores
If STM and LTM really are separate, recall should show their fingerprints. Murdock (1962) read participants lists of words and asked them to recall as many as possible in any order (free recall). Plot the chance of recalling a word against its position in the list and you get a U-shape — the serial position curve. Words from the start of the list are remembered well: the primacy effect, because those early words got the most rehearsal and made it into LTM. Words from the end are also remembered well: the recency effect, because they are still sitting in STM when recall begins. The words in the middle lose out both ways — too many to rehearse before the next ones arrive, and long gone from STM by the time recall starts.
The curve is powerful evidence because it does two things at once. It shows recall is not uniform, and it maps the two humps onto two different stores with two different explanations. If memory were a single store, there would be no reason for the beginning and the end of a list to be privileged over the middle.
Here is how to turn the curve into marks. A student writes: 'Murdock found people remembered words better.' That earns almost nothing — it names no effect and no store. A full answer reads: 'Murdock (1962) gave participants word lists to recall freely. Recall was best for the first words (the primacy effect), because they had been rehearsed most and transferred to long-term memory, and best for the last words (the recency effect), because they were still held in short-term memory at the point of recall. Middle words were recalled worst. This supports the multi-store model because the two peaks are explained by two separate stores.' Notice the shape of the mark-winning version: it names the study and date, describes the method in one clause, states both effects with the correct labels, gives the reason for each, and ends by linking the evidence back to the theory. That final linking sentence is the one weak answers leave out.
MechanismMemory as an active process — Bartlett, schema and effort after meaning
The multi-store model treats remembering as faithful storage and retrieval. Bartlett's Theory of Reconstructive Memory says the opposite: we do not store an exact copy and read it back — we store a gist and rebuild the rest each time, using what we already know. The engine of that rebuilding is the schema: a mental framework of expectations, built from past experience, about how the world usually goes. When you recall an event, your schemas fill the gaps, smooth the oddities and swap the unfamiliar for the familiar. That is why Bartlett's students turned canoes into boats and dropped the ghosts — their English schemas had no slot for those details, so the details quietly changed.
Bartlett called the driving force effort after meaning: the mind's insistence on making a confusing experience make sense, even at the cost of accuracy. This is not a laboratory curiosity. It is the reason eyewitness accounts drift, why rumours mutate as they pass along a chain, and why two people leave the same argument remembering different words. As evaluation, the theory has huge real-world reach and clear support from War of the Ghosts — but it is also criticised: Bartlett's instructions were vague, he did not always control how carefully people were trying, and reconstruction cannot explain the highly accurate, unchanging memories we sometimes have (a rehearsed poem, a multiplication table). The exam-ready judgement is that reconstruction explains meaningful, unfamiliar material far better than it explains rote-learned facts.
CasePutting it together — one topic, one 9-mark answer
The two big ideas in this topic are rivals, and AQA's favourite extended question sets them against each other: something like 'Discuss the multi-store model of memory. Refer to reconstructive memory in your answer.' A 9-mark 'discuss' is marked in three bands, and it rewards a specific balance — accurate description of the theory (AO1), then evaluation that uses the other theory and the studies as ammunition (AO3), finished with a judgement.
The scan you run on any memory essay is therefore always the same: describe the multi-store model accurately using coding, capacity and duration; support it with Murdock's serial position curve; then turn the reconstructive evidence into the critical half — Bartlett shows retrieval is not the clean readout the multi-store model assumes. A candidate who does only the first half caps at the middle band however fluent they are, because 'discuss' demands the evaluation. The worked example below shows exactly how much you need in each part.
A model middle-of-the-answer paragraph, annotated. Description (AO1): 'The multi-store model describes memory as three stores in sequence. Information enters the sensory register, and if attended to passes to short-term memory, which codes acoustically, holds about 7 items and lasts under 30 seconds. Rehearsal transfers it to long-term memory, which codes semantically and has unlimited capacity and duration.' — that is precise, quantified AO1; the numbers are what lift it above a vague description. Evaluation (AO3): 'A strength is that the serial position curve supports separate stores: Murdock found primacy (early words rehearsed into LTM) and recency (recent words still in STM) effects. However, the model is challenged by reconstructive memory. Bartlett's War of the Ghosts showed people alter memories to fit their schemas, which the multi-store model cannot explain because it treats retrieval as an exact readout rather than an active rebuild.' — notice each evaluation point is a claim plus evidence plus a consequence for the theory, never a bare opinion. Conclusion: 'Overall the model usefully explains the structure of memory and is backed by clear evidence, but it underplays how far everyday remembering is reconstructed rather than retrieved.' A one-sentence judgement that weighs both sides is what tips an answer into the top band.
VocabularyKey terms the mark scheme pays for
TrapsMisconceptions that cost marks
ExamWhat examiners want
Match your detail to the command word. 'Identify' or 'name' wants one or two words; 'describe' wants accurate AO1 with the numbers attached — a store is worth far more marks when you quote its coding, capacity and duration than when you just name it. For a 4-mark 'describe the multi-store model', the examiner is counting distinct, correct points, so give four: sensory register, STM, LTM, and the transfer between them.
The 9-mark questions are levelled (three bands) and split roughly between AO1 (accurate description) and AO3 (evaluation), with the top band reserved for answers that do both and reach a conclusion. 'Discuss' and 'evaluate' are not invitations to describe more — they demand strengths and weaknesses, each backed by a study (Murdock for the multi-store model, Bartlett for reconstruction) and each spelled out as a consequence for the theory, not a bare 'this is good'. Always finish with a one-sentence judgement that weighs the two sides.
On application ('source') questions, quote the scenario: if the source describes a witness whose account changed, name reconstructive memory and schema explicitly and tie them to the witness, because AO2 marks are only released when you connect the concept to the stem rather than defining it in the abstract.