Twelve rats were trained to discriminate 0.32 and 3.2 mg/kg s.c. midazolam from no drug under a three-lever, multiple trials drug discrimination procedure. In cumulative dose-response tests, midazolam s.c. (0.032-10 mg/kg) and i.p. (0.1-10 mg/kg) occasioned dose-dependent increases first in 0.32 mg/kg (low-dose) lever responding and then in 3.2 mg/kg (high-dose) lever responding. The benzodiazepines diazepam (0.032-18 mg/kg) and triazolam (0.0032-3.2 mg/kg) produced patterns of generalization similar to that of midazolam; however, chlordiazepoxide (0.1-32 mg/kg), lorazepam (0.032-10 mg/kg), flurazepam (0.01-10 mg/kg), bretazenil (0.01-32 mg/kg) and the imidazopyridazine zolpidem (0.032-3.2 mg/kg) dose-dependently occasioned > 80% responding on the low- but not the high-dose midazolam lever. Clonazepam (0.1-10 mg/kg) occasioned 0% responding on the high-dose lever, but also failed to occasion full generalization to the low-dose midazolam lever in 40% of the rats. Bretazenil has been well-characterized as a partial benzodiazepine agonist and zolpidem as benzodiazepine-receptor-subtype selective; the present results are consistent with their partial or selective agonist effects in those other paradigms. The differential effects of the classic 1,4 benzodiazepine agonists tested suggest that the discriminative stimulus effects of these other compounds may be more differentiable than previous drug discrimination studies have suggested. This three-choice drug discrimination procedure appears to be a useful model for studying relative intrinsic efficacies of this class of compounds.