Rats were trained to discriminate one of three doses of amphetamine (AM), 0.5, 1, or 2 mg/kg, from vehicle (VEH) in a two-lever, food-reinforced, drug-discrimination task. The purpose of the study was to investigate the nature of the shift of the dose-response curve and generalization to cocaine (COC) as a function of training dose. In order to preclude potential differences among the groups in stimulus control, the three training-dose groups were required to perform the discrimination at high and equivalent levels of accuracy. The shift of the dose-response functions to the right as a function of increasing training dose was not parallel. The slope decreased as training dose increased. There was a dose-dependent increase in AM lever responding to test doses of COC that tended to be affected by training dose. The results suggest that proper evaluation of training-dose effects requires that groups be trained to equivalent levels of stimulus control.