The multi-store model

Atkinson & Shiffrin (1968) proposed the multi-store memory model which is a structural model composing of 3 seperate stores with information passing between stores in a linear way. The 3 stores are the sensory model (SM), the short-term memory (STM) and long-term memory (LTM).

The SM has several stores called sensory registers (SR) with each processing information from a particular sense. The iconic register processes vision, the echoic register process sound and the haptic reigister process touch. The SR is not under cognitive control and is the first storage system for incoming information. Information recieved is raw and unprocessed and the SR has a large capacity however the duration of storage is milliseconds unless given focused attention in which case it will move to the STM.

Rehersal maintains information in the STM however it is still vulnerable to being forgotten due to limited duration or being displaced by new incoming information due to limited capacity. If rehersal and processed deep enough the information then passes to the LTM which has unlimited capacity and unlimited duration dependent on the level of processing of the information recieved. While the LTM encoding is mainly semantic and based on meaning, the STM encoding is acoustic with a capacity of 7+/-2 items and a duration of up to 18 seconds. Retreival of information from the LTM occurs through information passing back through the STM.

1 of 36

The multi-store memory model


2 of 36

Coding in the MSM

Coding is the way that information is changed so that it can be stored in memory. Information enters the brain via the senses. It is then stored in various codes (like a picture), acoustic codes (sounds) or semantic codes (the meaning of the experiences).

Sensory register/ sensory memory (SR)

  • Echoic store - auditory information
  • Iconic store - visual information
  • Haptic sotre - tactile information (touch)
  • Gustatory store - taste information
  • Olfactory store - smell information

Crowder (1993) found that SR only retains information in the iconic store for a few milliseconds, but for two to three for the echoic store, which suports the idea of sensory information being coded into different sensory stores (it also suggests they have different stores).

3 of 36

Coding in the STM - Badderly (1966)

Badderly (1966)

Aims: To assess whether coding in the STM is mainly acoustic or semantic (by meaning)


  • 75 participants were presented with 1 of 4 word lists repeated 4 times.
  • List A - acoustically similar words eg. cat, sat, mat.
  • List B - acoustically dissimilar eg. pit, huge, tall.
  • List C - semantically similar eg. big, huge, tall. 
  • List D - semantically dissimilar eg. hot, safe, fall.
  • To test coding, P's were given a list containing original words in the wrong order. They had to rearrange the words in the correct order; for STM. 
  • Procedure for the LTM was the same but with a 20 minute recall. 
4 of 36

Coding in the STM - Badderly (1966)


  • For STM, P's given list A performed the worst (recall 10%). They confused similar sounding words eg. 'cap' instead of 'cat'. Recall for the other lists was good, around 60-80%
  • LTM, participants with list C performed the worst, with a recall of 55. They confused similar meaning words 'big,' instead of 'huge'. Recall for other lists was good, 70-85%.


  • For STM, list A recalled the least efficiently 
  • For LTM, list C recalled the least efficiently 


  • Findings made 'cognitive sense'.
  • Small difference between semantically similar (64%) and semantically dissimilar (71%) lists. suggests there's also semantic coding in STM.
  • Lab study - shows causality, may lack ecological validity, replicable.
  • Lack of visual coding.
  • Independent group design. 
5 of 36

Capacity in the SR - Sperling (1960)

  • Flashed a 3x4 grid of letters onto a screen for 1/10 of a second and ask the P to recall the letters of one row. 
  • As the information would fade quickly he sounded different tones (high, medium, low) to indicate which one had been recalled (1st, 2nd and 3rd). 
  • Recall of letters in the indicated row was high, which suggests all the information was originally there, indicating that the capacity of the SR, especially for the iconic store, is quite large. 
  • Lacks ecological validity 
  • Can control all variables, lab study, can repeated.
  • Very much based on two stores: iconic and echoic, not the others.
6 of 36

Multi-store model (Strengths)

Glanzer et al demonstrated support for the STM and LTM being different stores. Participants were tasked with recalling word lists with earlier and later words more likely to be recalled and this was known as the primacy and recency effect. This primacy effect occurs as the first words are transferred to LTM while the recency effect occurs as the last words are still within the STM. Delays of 10 seconds or more before recall resulted in onlu a primancy effect with only LTM affected. This highlighted the difference in STM and LTM supporting the multi-store memory model theory.

A major strength of this model is that the models predictions around memory can be easily tested to verify whether it applies to human behaviour. The evidence supports the idea of STM and LTM being seperate types of memory and it has been verified through the use of PET scans and fMIR scans when participants have been doing seperate tasks related to STM and LTM. The prefrontal cortex is seen to relate to STM while the hippocampus is associated with LTM supporting the model's idea of different memory stores.

The model's predictions around memory can be easily tested to verify whether it applies to human behaviour. The evidence supports the idea of STM and LTM being seperate types of memory and it has been verified through the use of PET scans and fMRI scans.

7 of 36

Capacity in the STM

Jacobs (1887)

  • Tested STM capacity with a serial digit span method, where participants are presented with increasingly long lists of numbers or letters and have to recall them in the right order.
  • When P's fail on 50% of tasks they are judged to have reached their capacity.
  • Found that the capacity was between 7-9 numbers. 
  • Numbers may be easier to recall a there are only nine single-digit numbers (0-9), compared to 26 letters.
  • Criticism is ecological validity and mundane realism.
  • Criticism - age of study.

Miller (1956)

  • Reviewed research to find the capacity of STM to be 5-9 items, but that the chunks was the basic unit of STM.
8 of 36

Capacity of the LTM

Anokin (1973)

  • Estimated the number of possible neuroal connections in the human brain is followed by 10.5km of knots. He concluded 'no human yet can use the full capacity of their brain', suggesting the capacity of LTM is limitless.
9 of 36

Capacity of the MSM - Evaluation


  • Lab studies - high control, replicable
  • Links to evolution - Fagot and Cook (1996) showed that pigeons can memorise 1200 picture response associations. Babboons still hadn't reached that after 3 years of training, reaching 5000 associations. This suggests that an enlarged memory capacity has a survival value, which has been acted upon through natural selection. 


  • Lab studies - lacks ecological validity.
  • Other factors also influence STM eg. age. Jacobs also found that recall increased streadily with age; 8 year olds remember an average of 6.6 digits whereas the mean for 19 year olds was 8.6. This may be because of a gradual increase in brain capacity and/or it may be that people develop strategies to improve their digit span eg. chunking, as they get older. 
  • Conwar (2001) contradicts Miller, suggesting STM is more limited. He found that STM is likely to be limited to 4 chunks. This suggests the capacity may be smaller than previously thought. STM for visual info also found that 4 items was about the limit. This means that Miller's range is more appropriate. 
10 of 36

Duration of STM

Duration: the length of time information remains within storage.

Peterson and Peterson (1959)

  • Read nonsence trigrams to P's that didn't make up words, eg. FGT.
  • P's then had to count backwards in 3s to form a large 3 digit number for varying periods of time.
  • Found that 90% of trigrams were recalled correctly after 3 seconds, only 5% after 18 seconds - suggests STM has a capacity of between 20-30 seconds.


  • + Results likely to be reliable - Lab experiment; controlled variables.
  • - Results may be flawed due to methodology - different trigrams used in each trial - could lead to a difference between items, leading to decreased recall.
  • - Marsh et al found that if P's weren't expecting to recall info the STM was between 2 and 4 seconds - suggests the duration of STM is affected by time taken to process information. 
  • - Lacks mundane realism - don't encounter nonsense trigrams in everyday life.
11 of 36

Duration of LTM - Patrick et al (1975)

Duration potentially lasts a lifetime in the LTM.

Patrick et al (1975)

  • Showed 400 Ps aged 17-74 a set of photos & a list of names, some of which were ex-school friends, and asked them to identify ex-school friends. 
  • Those who'd left high school in the last 15 years identified 90% of faces and names. 
  • Those who'd left 49 years before identified 80% of names and 70% of faces, suggesting memory for faces is long lasting. 
  • Sometimes info in LTM appears to be lost but there's a problem of access to the info rather than it not being there.
  • No guarantee that some people have recently looked through their yearbook.
  • Some faces may have had more of an impact.
  • Higher ecological validity.
  • There was a photo recognition group for matching names and faces.
  • There was a name recognition recall group for people to remember names through photos. 
12 of 36

Duration of LTM - Patrick et al (1975)


  • Some people can remember info for a lifetime.
  • Very LTM is better for recognition rather than recall.


  • + Field experiment - high ecological validity
  • + Has meaningfulness and relevance - showed better recall than other studies. 
  • -  It is unclear whether the drop-off in accuracy after 49 years reflects the limits of duration or a more general decline of memory with age.
13 of 36

Multi-store model (Weaknesses)

The MSM can be argued to be oversimplifying memory structures and processes. Shallice et al highlighted this with a case study of KF who suffered brain damage resulting in difficulty with verbal information in STM but normal ability with visual information. This highlights how STM is not a single store as the MSM suggests. LTM may not be a single store either as Schachter et al proposed 4 different types of LTM stores; semantic, episodic, procedural and PRS.

Clive Wearing lost his episodic memory but not procedural, again suggesting more than one type of LTM store.

14 of 36

Long term memory

LTM is believed to be divided into two main types of memory. The first is Declarative memory which is a type of concious memory also known as 'knowing that' memory which helps us recall facts. This is subdivided into episodic memory and semanti memory. The other is not conciously recalled and is known as 'knowing how' memory, or procedural memory. This helps us recall procedures such as how to tie our shoes, cycle or swim.


15 of 36

Episodic memory

Consists of memories such as our thoughts or experiences we have had and our personal recollection of them. Memories that are episodic are usually based on events that occur in people's lives however over time they move to semantic memory as the event association diminishes and the moemory becomes knowledge based. The strength of episodic memories is determined by the emotions present at the time the memory is being coded. Traumatic life events may be recalled better due to the strong emotional attachment they have. The pre-frontal cortex of the brain is linked to the initial coding of episodic memories and consolidation and storage associated with the neocortex. 

16 of 36

Semantic memory

Contains knowledge, facts, concepts and meanings the individual has learnt. Semantic memory may also relate to how certain objects work, their functions, appropriate behaviour in situations or abstract concepts such as language or maths. The strength of semantic memory is positively correlated with the strength of processing that occurs when coding with semantic memories lasting longer than episodic memories. Semantic LTM is linked to episodic LTM with semantic memories formed based on experiences that occur. Therefore episodic memory underpins semantic memory with episodic based eperiences moving over to semantic memory over time. Semantic coding is mainly associated with the frontal and temporal lobes with opinion on semantic LTM being mixed; some argue the hippocampus is involved while others believe several parts of the brain play a role. 

17 of 36

Procedural memory

Is skill based memory and docused on recalling how to so something and does not require concious thought. Procedural memories are usually learnt through repetition and practice. Language is believed to be a procedural memory as it helps individuals speak using the correct grammar and syntax without having to conciously give this thought. Procedural LTM is linked to the neocortex brain areas within the primary motor cortex, cerebellim and prefrontal cortex. This is different to declarative memory stores as it does not rely on the hippocampus to function. 

18 of 36

LTM - Strengths

Brain scans have shown 3 distinct areas being active with the hippocampus and other parts of the temporal lobe such as the frontal lobe associated with episodic memory. Semantic memory has been associated with activity in the temporal lobe while procedural memory associated with the cerebellum and motor cortex. This supports theories for 3 distinct stores of LTM as theories suggest.

Case studies such as HM (Milner 1962) support the case for procedural and declarative memory stores being distinctively different. HM could not form episodeic or semantic memories due to the distinction of the hippocampus and temporal lobes however he was able to form procedural memory through learning how to draw figures by looking at their reflection. However, he could not recollect how he had learnt this skill, supporting the case for different stores between semantic and episodic memory. 

Vicari et al (2007) - Case study of CL who suffered brain damage after the removal of a tumor found deficiencies in ability to create new episodic memories. However, she was still able to create semantic memories supporting the theory that they are seperate stores. 

19 of 36

LTM - Weaknesses

A lot of research for LTM comes from case studies. These are based on a single individual therefore making it difficult to generalise the findings to the wider population as it deficits in memory may be unique to this one person. 

Another weakness into theories of LTM is the lack of research into brain areas that are involved in procedural memory. Case studies of individuals with brain damage that affects procedural memory but not declarative memory is needed ot understand this bettwe however such cases are extremely rare. Therefore we cannot conclusively say the procedural memory store is fully understood with any detail to generalise such a theory.

20 of 36

The Working Memory Model

Baddeley's WMM replaced the idea of a unitary STM store and suggested a system involving multiple stores consisting of active processing and ST storage of information. In this model STM is an actice processor (working memory) which the central executive (CE) 'attends to and works on' either speech based information recieved from the articulatory-phonological loop or visually coded information recieved from the visuospatial sketchpad.


21 of 36

The Central Executive

Is the main component and coordinates the other 'slave systems' and ensures they don't go astray. The slave systems (PL, VSS and EB) can also be used as temporary storage systems to free up capacity within the CE to deal with other demanding tasks. Capacity for the is very limited anf it revieves information from the senses or from LTM. It is also involved in directing 'attention' and resources towards particular tasks. 

22 of 36

The Articulatory-Phonological Loop

The PL processes speech based information preserving this order within the phonological store which acts like an 'inner ear'. The articulatory control process (inner voice) is linked to speeech production and is used to reherse and store verbal information from the phonological store through a form of maintenance rehersal.  Again this has limited capacity determined by the amount of information which can be spoken out loud in 2 seconds. Confusions often occur with similar sounding words as it is an acoustic store. 

23 of 36

The Visuospatial Sketchpad

Processes visual information through the senses (eyes) or LTM on what things look like, patterns of recognition and spatial information consisting of the relationship between things and perception of movement. Logie (1995) suggested the VSS could be further sub-divided into a visual cache (CV) which stores visual material on colour and form and an inner-scribe (IS) which deals with spatial relationships, rehersal and the transfer of information from the visual cache to the CE.

24 of 36

The Episodic Buffer

While the PL and VSS deal with the processing and temporary storage of specific types of information, the EB works as an extra buffer that integrates information from all 3 main systems as well as LTM. The CE has no storage capacity of its own and Beddeley reaslised the model required a general store to explain why some amnesia patients with no longterm recall could recall information immediately. This suggested a time-sequencing recordig events as they happen and transferring this information into the LT capacity.

25 of 36

WMM - Strengths

Baddeley demonstrated the existance of the VSS when Ps were given the task of tracking a moving light with a pointer. While doing this they were tasked with one of two other tasks: one to describe all the angles on the letter F and another to perforem a verbal task. Describing the angles was difficult as both tashs competed for the limited resources of the VSS but not the verbal tasl as that involved two slave systems. This demonstrated the limited capacity of the VSS but also how it differs from the PL which is responsible for auditory tasks. 

Klauer & Zhao provide support for the visual cache and inner scribe. They found more interference occured between two visual tasks compared to a visual and spatial task suggesting both were seperate components with the visual cache dealing with colour and form and the inner scribe dealing with spatial relationships. PET scans also support these findings with brain activation apparent in the left hemisphere when doing visual tasks and right hemisphere activity when doing spatial tasks which supports the idea that the VSS is further subdivided into a seperate visual cache and inner-scribe.

D'Espositio et al (1995) used fMRI scans to find that the pre-frontal cortex was activated when verbal and spatial tasks were performed simultaneously, but not seperately, implying that the brain area is linked to the CE.

26 of 36

WWM - Weaknesses

There is little known abou tthe main component, the CE or how it works and there is evidence suggesting it is not unitary. Critics argue the CE may not be a single elemet and Eslinger et al highlighted this with one patient EVR who had a cerebral tumor removed. While he performed well on reasoning tasks suggesting his CE was functional, he struggled with poor decision making skills suggesting some elements of his CE were partly damaged. This suggests there may be other components to the CE which the WMM is unable to explain due to it being over-simplified in its theory.

Case studies - the process of brain injury is traumatic, which affect attention and performance of tasks.

WMM concerned itself with only STM, so is not a completed or a proper model.

Much of the evidence for VSS and the OL have often relied on dual task techniques. The problem with these is that they lack external validity and realism due to the artificial setup and they are not tasks people  would usually do in everyday life. Therefore studies that have tested for the WMM lack mundane realism. 

27 of 36

Explanations for forgetting - PI

Interference theory argues forgetting occurs die to two memories competing and being affected by past memories or possible future learning. The more similar the two memories are the more interference it causes as the two memories become confused with one another. 

Proactive interference: forgetting occurs when older memories, already stores, disrupt the recall of newer memories. Degree of forgetting is worse when memories are similar. 

Keppel and underwood (1962) demonstrated proactive interference. Ps were tasked with recalling cosonant trigrams after varying intervals where they were tasked with counting backwards in threes. Forgetting increased after each interval however little forgetting occured at the start. PI can explain this as earlier consonants entered the LTM and thus interfered with the formation of new memories.

28 of 36

Explanations for forgetting - RI

Retroactive interference occurs when newer memories disrupt the recall of older memories already stored. Forgetting is worse when memories are similar. Ceraso (1967) suggested one possible explanation for RI was there was no actual loss of information but merely the wrong information was accessed as it had been moved. Unlearning theory however suggests the new learning replaces the old learning.

Muller (1900) identified RI through a study where participants tasked with learning a list of syllables are given an intervening task between exposure to the syllables and recall. The intervening task (describing paintings) produced RI with Ps struggling to recall their lists.

29 of 36

Interference theory - Strengths

McDonald et al (1931) experimented with Ps giving them lists of adjectives to remember. After learning list A they were given list B and tasked with learning this. Recall was found to be poorest when list B was a list of synonyms of list A supporting the case for confusion to occur between the two memories as interference theory states. 

Offers real world application. Danaher (2008) found when people were exposed to adverts from competing brands withing a short time frame, Ps struggled to recognise the brands or their message. Considering the millions spent on advertising this presents a big problem but also provides marketers practical ways to overcome this. By ensuring adverts are spaced significantly far apart from the airing of rival brands or by repeating more on one day rather than over the week with rival brands, this can help avoid dilution of adverts.

30 of 36

Interference theory - Weaknesses

Interference effects are more evident in lab settings using various memory based tasks. These set-ups lack ecological valifity and also mundane realism as the tasks are rarely indicitive of what people would experience in real-life situations. Therefore it makes it difficult to generalise their findings externally beyond the lab settings or understand exactly how much day-to-day forgetting can be credited to interference or even forgetting in general. Anderson (2000) believed interference did play a role in forgetting but it was difficult to understand exactly how much. 

Kane et al (2000) found individuals with bigger working memory spans were less susceptible to PI when testing recall using three word lists compared to individuals deemed to have less working memory spans. It is unclear whether those woth greater working memory spans have achieved this either through more practice in some form but it highlights how interference theories cannot be fully generalised to everyone.

Interference theory is it only explains forgetting when information is similar and can not explain why forgetting occurs in the majority of real life situations. Also forgetting due to similarities doesn't happen that often either suggesting it is only one part of a bigger explanation and over-simplified. The fact that there is significant research support for cue-dependent forgetting suggests other explanations or processes must be at which and interference theory cannot explain everything. 

31 of 36

Explanations for forgetting - Retrieval failure

Retrieval failure theories argue forgetting from the LTM is caused by failing to access the memory due to insufficient clues or cues to aid recall rather than it being unavailable. Similar to a labelling system on a filing system, the cues act as markers to aid effectiveness depends on the number of items associated with it fewer items leading to a more effective cue.

Tulving (1973) called this this the encoding-specificity principle; where recollection is affected if the context of recall is different to what is was when the memory coded. He suggested that memory recall is most effective when information which was present at the time of encoding is available during retrieval. There are two main tyoes of cue-dependent forgetting; context dependent failure and state dependent failure.

32 of 36

Context dependent failure

May rely on external environmental retrieval cues being similar to when the information was encoded to aid recall, eg. being in the same room where you learnt the answers to a test and then taking the test in this room. This would result in greater recall than being in a different room. Environmental context such as being at a particular place can trigger retrieval as can particular sights or sounds if they are experienced stongly enouch during encoding. 

Abernathy (1940) found that after participants had learnt various material, they showed greater difficulty with recall when they were tested by an unfamiliar teacher in an infamiliar room compared to a familiar teacher and familiar room. This shows support for the importance of context aiding the memory retrieval process. 

33 of 36

State dependent failure

Would occur when the internal state of the person is different to when the information was encoded. This may be down to feeling a different emotion for example and trying to remember something when you were happy while you are feeling sad. Therefore internal states can also act as retrieval cues.

Support for state-dependent failure comes from a study by Overton (1972). Participants learnt material either drunk or sober and found participants struggles with recall more when trying the information in a state that is different to the time of encoding. For example, trying to recall information sober when it was learnt drunk (and vice versa). This provides support for state dependent faolure as an explanation for forgetting.

34 of 36

Retrieval failure - Strengths

Research into retrieval failure due to cue dependent forgetting has real world applications particularly in the search for missing people and reconstructing the last known whereabouts. This was used in to aid in the conviction of Danielle Jones' killer as a reconstruction in 2001 prompted witnesses to recall her arguing with her uncle through witness testimony. Therefore understanding how cues affect recall can help us develop ways to improve memory for the benefit of society.

Baddeley (1975) found evidence for cue dependent learning and how context cue aided retrieval. Divers tasked with learning material either on dry land or while underwater were found to have poorer recall when they were tested on retreival in a context that differed from where encoding and learning happened. For example, testing them for material they learnt underwater while on land resulted in poorer retrieval than if they were tested underwater. The sam ewas true vice versa too with better recall shown when the learning context remained the same as encoding. This supported cue-dependent failure however this was during free recall only. When given a recognition test and asked to say whether the item on the list was in the learning list or not, context based failure effects were not observed showing how cue dependancy can't explain all forms of forgetting.

35 of 36

Retrieval failure - Weaknesses

Many studies into retrieval failure due to cue dependent forgetting are based in the lab and lack ecological validity and mundane realism as they are not indicitive or real world environments or situations of forgetting. Also such explanations are not able to explain why retrieval failure cannot be explained with due dependent forgetting for activities such as riding a bike suggesting retrieval failure as a theory for forgetting is oversimplified and incomplete. 

Research into state dependent failure such as Overtons (1972) study raise ethical concerns as they encourage people to become drunk and under the influence of substances which can lead to injury or death even by accident. Also the level of engagement from Ps when under the influence of alcohol may not necessarily be genuine due to the way it affects people's willingness to give honest answers. Some Ps may have deliberately done poorly in some situations or try harder in others due to how alcohol affects people in unpredictable ways.

Baddeley (1997) criticised the encoding specificity principle as impossibel to test and verify for certain making it unscientific. If a cue aids retrieval them it could be argued to have been encoded in the memory however if it does not then it could be argued that it wasn't encoded in memory as a cue. The fact that it is impossible to test for an item as having been encoded or not means we cannot fully test the encoding specificity principle.  

36 of 36


No comments have yet been made

Similar Psychology resources:

See all Psychology resources »See all Memory resources »