Multi-Store Model

The three stores are: (1) the sensory register — holds raw sensory information for a fraction of a second (iconic memory ~0.5 s; echoic memory ~2–4 s); very large capacity. (2) Short-term memory (STM) — holds information that has been attended to for approximately 15–30 seconds without rehearsal; capacity ~7 ± 2 chunks (Miller, 1956); uses predominantly acoustic coding. (3) Long-term memory (LTM) — effectively unlimited capacity and potentially permanent duration; uses predominantly semantic (meaning-based) coding.

Short-term memory primarily uses acoustic (sound-based) coding. Conrad (1964) showed that recall errors in STM clustered around acoustically similar letters — even when those letters were presented visually — indicating the material had been recoded into phonological form. Long-term memory primarily uses semantic coding. Baddeley (1966) showed that semantically similar words cause more interference in LTM recall than acoustically similar words. This coding distinction is a key piece of evidence for the STM/LTM separation proposed by the multi-store model.

Key evidence includes: (1) The serial position effect — better recall for early items (primacy, attributed to LTM transfer via rehearsal) and final items (recency, attributed to STM). Delay between learning and recall selectively abolishes the recency effect, consistent with STM decay. (2) Patient H.M., who had intact STM but could not form new LTM — supporting their separation. (3) Sperling's (1960) iconic memory work establishing the sensory register. (4) Peterson and Peterson's (1959) decay study establishing STM's brief duration.

The major criticisms are: (1) STM is not unitary — patient K.F. had severely impaired STM but intact LTM formation, contradicting the serial gateway. (2) Rehearsal does not equal LTM transfer — levels of processing research (Craik & Lockhart, 1972) showed that deep, semantic processing produces better retention than extended maintenance rehearsal. (3) The model ignores the active processing role of STM, treating it as a passive temporary buffer. These criticisms were addressed by Baddeley and Hitch's (1974) working memory model, which replaced the unitary STM with a multicomponent system.