The impact of training sequence on discrimination of a mixture of two drugs was investigated with five groups of rats (n = 10). In phase I, two groups were trained according to conventional two-lever, operant drug discrimination protocols with food reinforcement; one of these groups was trained with nicotine (0.4mg/kg) and the other group was trained with midazolam (0.15mg/kg). The three remaining groups served as controls and were subjected to 'sham' training in which administrations of saline, nicotine or midazolam were unrelated to contingencies of reinforcement. After completion of phase I (40 sessions), all five groups were trained to discriminate a mixture of nicotine (0.4mg/kg) plus midazolam (0.15mg/kg) from saline (phase II). Any differences between the groups in their performance during phase II could, therefore, be attributed to their different histories in phase I. During phase II, all groups discriminated the mixture from saline with similar accuracy (89-94% drug-appropriate responding after mixture as compared with 2-7% after saline). In the three groups of rats subjected to 'sham' training in phase I, there was partial generalization to both nicotine (45-53%) and midazolam (39-40%), each of which therefore contributed about equally to stimulus control by the mixture. In rats that were initially trained to discriminate nicotine, midazolam had acquired little stimulus control over behaviour (9%) and discrimination of the mixture was attributable largely to the nicotine (87%). Conversely, in rats that were initially trained to discriminate midazolam, nicotine contributed 3% and midazolam 76% to stimulus control by the mixture. These powerful, persistent effects of training sequence were interpreted as examples of associative blocking demonstrated with the interoceptive stimuli produced by psychoactive drugs.