The drug discrimination paradigm, which is sometimes considered to provide an animal model of human subjective effects, has been adapted for use with human subjects. This procedure provides a direct measure of stimulus similarity between the test drug and a known standard drug, information that may be of value in predicting abuse liability of novel compounds. The present paper reviews the general methods used in human drug discrimination studies, some of the drugs that have been tested, factors to consider in designing drug discrimination studies in humans, and evidence concerning the utility of the procedure in abuse liability assessment. Further research is needed to validate this procedure and to determine the conditions under which it has the greatest value in abuse liability assessment.