In rats trained to discriminate 0.04 mg/kg fentanyl from saline, phencyclidine (PCP) and the PCP-type drugs ketamine and (+/- )-5-methyl-10,11-dihydroxy-5H-dibenzo(a,d)cyclohepten-5,10-imine produced effects that are usually referred to as partial generalization. Partial generalization could conceivably result from low efficacy actions at the receptor mediating the discriminative stimulus effects of the training drug. The PCP-type drugs produced maximum percentages of drug lever (DL) selection intermediate between those produced by the training conditions, but their curves relating dose to percentage of DL selection were not shallower than that of fentanyl. The PCP-type drugs decreased DL selection produced by the training dose of fentanyl, but there was no relationship between these antagonist effects and the DL selection produced by the PCP-type drugs when given alone. Naltrexone antagonized DL selection produced by fentanyl, but not that produced by the PCP-type drugs. The potency order of the PCP-type drugs to produce DL selection was in agreement with their relative affinities for PCP receptors, but not with those for morphine receptors. The intermediate levels of DL selection produced by the PCP-type drugs were associated with increased lever selection latencies and increased responding on the nonselected lever; this pattern of effects resembled the behavior of animals that had not yet acquired the discrimination. The results suggest that PCP-type drugs produce intermediate levels of drug-appropriate responding in fentanyl-trained rats through mechanisms involving not opioid receptors and partial generalization, but involving PCP receptors and performance deficits conceivably resulting from state dependency. Thus, the results stress the importance of a pharmacological and behavioral analysis of intermediate responding in drug discrimination to examine its validity as a measure of efficacy and of stimulus similarity.