In experiment 1, rats (n = 12) were trained to discriminate the benzodiazepine (BDZ) compound chlordiazepoxide (CDP, 20 mg/kg, IP) from saline in a two-lever food-reinforced procedure, and subsequently were tested for stimulus control with different doses of CDP, Ro 15-1788 (a proposed BDZ receptor antagonist) and Ro 15-1788 plus 20 mg/kg CDP. Ro 15-1788 (0.63-40 mg/kg) dose-dependently antagonized CDP, and induced predominantly saline appropriate responding when administered alone. Thereafter, the same rats were retrained by progressively decreasing the training dose, to discriminate 2.5 mg/kg CDP from saline, and were tested again with the same compounds. Ro 15-1788 (0.16-40 mg/kg) now failed to antagonize CDP (2.5 mg/kg) and increased the percentage of drug-appropriate responding in a dose-related manner when administered alone. In experiment 2, separate groups of rats (n = 10) were similarly trained to discriminate either 15 or 3 mg/kg CDP from saline. Tests with CDP, Ro 15-1788 and Ro 15-1788 plus CDP (either 15 or 3 mg/kg) yielded similar results to experiment 1, suggesting that the training dose effects on generalization and antagonism of Ro 15-1788 were not affected by the manner in which the lower CDP dose acquired drug stimulus control. It is concluded that mixed agonist-antagonist properties are apparent after variations of the BDZ training dose in a drug discrimination procedure.