Two experiments were performed that involved an initial word-identification task in which the acoustic signal was degraded by either 50% or 60% compression of the recorded words. A control group was tested at the original recording rate (0% compression). The compression manipulation was successful in increasing the number of identification errors. During the second stage of the experiments, words were read at a normal rate. In Experiment 1, the subjects indicated whether each word had appeared on the prior list, and in Experiment 2, the subjects estimated the number of times each word had been read on the prior list. The false-negative rate decreased and frequency ratings increased as a function of the number of prior presentations (one, two, or three) of the target words. The most interesting result was that the effect of repetitions was present even for words that had been misperceived.