The improvement in word intelligibility associated with successive presentations of a word in noise was examined with recorded and with independent samples of the speech and/or its background noise. The listener's criterion for termination of a trial was shown to be a crucial determinant of the gain of intelligibility with successive presentations. The improvement in intelligibility was greater for independent samples of speech and noise than for successive presentations of a single recorded sample. The observed improvement, however, was less than predicted for the ideal observer with perfect memory by the theory of signal detectability. Indexes of response discriminability and signal discriminability are briefly discussed.