Estimates of the likelihood that a drug will be abused have generally been based on the subjective effects engendered by that drug. With the development of standardized subjective effects questionnaires in the 1960s and 1970s, researchers have been able to carefully evaluate self-reported effects of drugs, generally making measures before and repeatedly after administration of a single dose of drug. The use of multiple doses under controlled laboratory conditions in which physiological measures are also taken, and both the investigator and the subject are blind to the dose administered, has been suggested as most likely to yield useful data about the abuse liability of a test compound. Although questions remain about the specific subjective effects measures to be used, there has been general agreement among researchers in this area that scores on scales from the Profile of Mood States, Addiction Research Center Inventory, and Visual Analog Scales which include measures of 'high' or 'liking' all provide predictive utility. The addition of a measure of actual drug-taking to this predictive model appears to provide important information about the conditions under which these two behaviors (self-reported effects and drug self-administration) vary, and strengthens the model substantially.