In 3 experiments, we examined Perruchet, Cleeremans, and Destrebecqz's (2006) double dissociation of cued reaction time (RT) and target expectancy. In this design, participants receive a tone on every trial and are required to respond as quickly as possible to a square presented on 50% of those trials (a partial reinforcement schedule). Participants are faster to respond to the square following many recent tone-square pairings and slower to respond following many tone-alone presentations. Of importance, expectancy of the square is highest when performance on the RT task is poorest--following many tone-alone trials. This finding suggests that RT performance is determined by the strength of a tone-square link and that this link is the product of a non-expectancy-based learning mechanism. The present experiments, however, provide evidence that the speeded RTs are not the consequence of the strengthening and weakening of a tone-square link. Thus, the RT Perruchet effect does not provide evidence for a non-expectancy-based link-formation mechanism. (Contains 4 figures.)