An analysis of previous studies of overtraining and reversal learning in rats suggests that overtraining has only facilitated reversal when the discrimination has been a relatively difficult one and the reward used has been relatively large. The results of three experiments support this conclusion. In the first, overtraining had no significant effect on the reversal of an easy position problem, whether a small or a large reward was used; while in the second, using the same procedures and large reward, overtraining facilitated reversal of a visual problem. In the third experiment, rats were trained on an easy or on a hard visual problem with either a small or large reward; only /Ss trained on the hard problem with the large reward reversed faster after overtraining. A two-stage attentional model is better able to explain these results than are either frustration theory or a combination of attention and frustration theories.