Three studies were conducted to assess the effects of subliminal reinforcements on learning. In the first two, Ss were given a discrimination task in which five geometric forms, repeated over 100 trials, were to be assigned to one of two categories. The categories were unbalanced; four geometric forms comprised one category, the remaining form the other. Response was required on each trial. Immediately after each response, the appropriate reinforcing word, “Right” or “Wrong,” was flashed at a subliminal brightness-contrast In Exp. I, under low motivation (without money incentives), Ss showed no learning of the correct discrimination, nor any evidence of probability learning with respect to relative frequency of stimulus categories. In Exp. II, the above procedure was replicated with money as the incentive. There again was no evidence of discrimination learning, i.e., acquisition of the correct response. There was, however, a significant linear trend ( p < .05) in the proportion of responses made to the more frequent stimulus category; Ss showed an increasing tendency to “match” the relative frequency of their two classes of response with the corresponding two stimulus classes. In Exp. III, Ss who were motivated by a money incentive attempted to guess whether E was thinking of an odd or an even number. Following each response, Ss were reinforced by tachistoscopic presentation of the word “Right” or “Wrong,” at time intervals which were too brief to permit recognition; half of the Ss were positively reinforced for emitting the response “Odd,” and half for the response “Even.” After 100 learning trials had been completed, the reinforcement contingencies were switched for an additional 20 trials, e.g., Ss who had been reinforced for “Odd” were now reinforced for “Even.” Ss in Exp. III showed no evidence of probability learning. Some possible explanations for the conflicting results of Exps. II and III were discussed.