A model of the response process based on signal-detection theory (Tanner & Swets, 1954) was used to analyse PI effects in three different tests of short-term memory. The focus of interest was the effect of interference on the variable d', when the latter is used as a measure of retention theoretically independent of response bias. Three experiments used varying conditions of interference and varying types of recall test: it was found that d' was unchanged, or slightly raised, for an item which had previously been learned in the same serial position. If, however, the item had occurred previously without being memorized, the effect was to lower d', i.e. to impede learning. A response bias existed in all experiments towards items previously presented.