In drug discrimination studies benzodiazepine-trained animals have typically responded on the drug lever when tested with barbiturates. In a recent study, greater specificity appeared to be shown when lorazepam was used as a training drug. The generality and limits of this finding were explored in the present set of experiments. The asymmetrical cross-generalization found in lorazepam- and pentobarbital-trained baboons was replicated in rats and was shown not to be a function of either lorazepam (0.1., 0.32, or 1.0 mg/kg) or pentobarbital (10 or 25 mg/kg) training dose (i.e., pentobarbital-trained rats responded on the drug lever in tests with lorazepam, but lorazepam-trained rats did not show comparable pentobarbital generalization). In the next experiment, groups of rats were trained to discriminate chlordiazepoxide (10 mg/kg), triazolam (0.1 mg/kg), or diazepam (1.0 mg/kg). Generalization to both lorazepam and pentobarbital was shown by these rats. Finally after daily pentobarbital administration, lorazepam-trained rats made a sufficient number of responses after high pentobarbital doses to permit extension of the range of pentobarbital doses tested. Pentobarbital generalization increased, but still did not occur in all rats and was unreliable in successive tests in the same rats. These results suggest less homogeneity in the discriminative stimulus effects of "depressant drugs" than generally has been recognized.