To learn what maintains the frequency of shuttle box avoidance responses, male rats from the Berkeley S1 strain, after 200 trials of standard discriminative avoidance training, were given 100 additional trials under one of four different conditions. Responding at the maximum rate was maintained when animals performed under the training conditions or when responses continued to terminate the warning signal immediately, even though shock was never given for failing to respond. In contrast, avoidance responding was reduced markedly if, and only if, trials were given in which the signal ceased to terminate immediately (i.e. it shut off either well before or well after a response). This decrement occurred even though avoidance responses continued to avert shock. Thus, under the conditions of this experiment prompt signal offset was both necessary and sufficient to maintain the occurrence of well-established shuttle box avoidance responses.