To assess the discriminative stimulus properties of mixed agonist-antagonist opioids in humans, postaddict volunteers were trained in a three-choice drug discrimination procedure to discriminate among the effects of saline (4 ml i.m.), hydromorphone (3 mg i.m.) and butorphanol (6 mg i.m.). Subjects earned monetary reinforcement by correctly identifying the training drugs by letter code. Other subjective, behavioral and physiological measures were concurrently collected. After training, generalization curves for hydromorphone, butorphanol, pentazocine, nalbuphine and buprenorphine were determined. In generalization testing, both hydromorphone and butorphanol produced dose-related increases in hydromorphone-appropriate and butorphanol-appropriate responses, respectively, and other characteristic subjective effect measures. Nalbuphine produced dose-related increases in discrimination as butorphanol and in those subjective effect measures increased by butorphanol. Buprenorphine produced dose-related increases in discrimination as hydromorphone and in those subjective effect measures increased by hydromorphone. Pentazocine was not consistently discriminated as either butorphanol or hydromorphone. These results differ from those of previous discrimination studies using similar methods but different training drugs. Compared to a previous study in which pentazocine served as the kappa-like training drug, the use as a training drug of the more pharmacologically specific kappa-like drug butorphanol permitted greater differentiation among test drugs and yielded discrimination results more consistent with other pharmacological evidence (buprenorphine being mu-like and nalbuphine being kappa-like). There was a close relationship between results of the discrimination measures and the subjective effect measures.