According to signal-detection theory the response is a function of both sensitivity and the criterion; this experiment tested for criterion shifts in short-term memory (STM). Each list consisted of 5 A-B paired associates followed by a probe (A or B) for 1 of the pairs. 4 Ss were tested intensively, and recall was supplemented with confidence ratings and latency measures. Analysis of the results showed that the strength of the evoked response (as measured by d') was invariant over serial position but the criterion (as measured by beta) became stricter as retention interval increased. Thus the predicted changes did occur; these findings raise doubts about conventional error analyses in studies of human verbal learning and suggest caution in extrapolating from probability measures to strength measures. (18 ref.) (PsycINFO Database Record (c) 2006 APA, all rights reserved)