Pigeons were trained in a symbolic matching-to-sample task in which food (access to grain) and no-food (blackout) samples were used. Responding to red test stimuli was reinforced on food sample trials, and responding to green test stimuli was reinforced on no-food sample trials. Samples of food and no food were rendered "expected" or "surprising" during subsequent testing by preceding samples with a previously established signal for either food (CS+) or no food (CS-). CS+ preceded food samples and CS- preceded no-food samples on expected trials, and CS- preceded food samples and CS+ preceded no-food samples on surprising trials. The findings suggest that enhanced retention on surprising trials is apparent from the outset of testing to the extent that (a) surprising and expected trials are not markedly dissimilar to the trials of training, and (b) control over test responding by the CSs presented on expected and surprising trials is minimized. It is concluded that surprising samples receive enhanced postperceptual processing (i.e., rehearsal) and that the mechanism involved is that of retrieval-generated priming of short-term memory.