It seems highly unlikely that STM only has new information in it - since we can use chunking that we must accept that LTM is being used to organise information. De Groot(1966) looked at chess players by showing them a chessboard with pieces on. He found that grandmasters were better than beginners at remembering the pieces and their locations only if the pieces were in positions that you would find in a real game. If the pieces were just randomly around the board then beginners were just as good at remembering them. Clearly there is a lot of top-down stuff going on to organise the info.
Craik and Lockhart have also demonstrated that it is not rehearsal that is important but the kind of rehearsal and that elaborative rehearsal is the best way to have deeper encoding of material and therefore more permanent.
The properties of sensory register are based on Sperling's (1962) findings from the part-whole procedure. Selected information from the sensory store then enters short-term store, and is held in a symbolic code. A key tenet of the theory is that the code is auditory-visual-linguistic in nature, although short-term stores for different modalities (e.g., kinaesthetic) are not ruled out. The assumption that information is stored in an auditory-verbal-linguistic code is based on Conrad's (1964) finding that errors in short-term memory tasks tend to be based on phonological rather than semantic confusions. Information from the long-term store can also enter the short-term store. Information in the short-term store is lost within about 30 s if not maintained through rehearsal (i.e., repetition that increases length of time information held in short-term store), and the strength of a long-term representation is determined by amount of time information spends in the short-term store. Rehearsal is viewed as a process that regenerates the short-term store trace, and the capacity of the short-term store is equal to the amount of information that can be rehearsed. However, Atkinson and Shiffrin also recognise that information may be transferred to the long-term store without rehearsal, and also that a long-term trace will be stronger if other control processes are used, for example, coding. The idea of a rehearsal buffer--a bin with a fixed number of slots--is also discussed, both as a rehearsal strategy and possibly as a structural feature of the short-term store. Decay is the mechanism responsible for information loss. Another key tenet is that information is lost through decay, as opposed to interference. Long-term memory is, by contrast, permanent, and therefore different mechanisms are responsible for information loss in the short-term and long-term stores. Control processes associated with the long-term store are responsible for transfer from STS to LTS and for search/retrieval.
Of course, later theorists (e.g., Shiffrin & Schneider, 1977) assumed that such automatic "transfer" is obligatory.
Atkinson and Shiffrin distinguish between the term’s short-term store and long-term store, on the one hand, and short-term memory and long-term memory, on the other hand. The latter terms refer to memory paradigms, whereas the formers are theoretical constructs. Both memory stores are involved in short-term memory and long-term memory experiments.