The drinking histories of 102 patients were elicited by both a computer and a nurse, with 48 of the patients also being interviewed by a psychiatrist. The computer agreed as closely with the nurse and the psychiatrist as they did with each other. For all three, there was better agreement for screening interviews whose questions were dichotomous (i.e. yes/no) than for interval data (how much, how often). Nevertheless, the agreement for alcohol volume was as satisfactory as can be expected from the literature on self reported measures. 3 figures. 27 references. (Author abstract modified)