Drinking episodes during the treatment (relapses or lapses) of alcohol-dependent patients is predicted from clinical ratings of patients and individual background data such as alcohol drinking history and social status. The probability of these relapses (or lapses) is determined up to three days in advance using a logistic regression procedure. The study group consisted of 33 male alcohol-dependent persons, who participated in a treatment program. Clinical ratings were performed three times a week by a trained person during a visit to the clinic. The questionnaire contained 23 different items about irritation, craving for alcohol, sleep disturbances, etc. The relapses were either self-reported or detected by a biochemical marker in a urine sample that was taken daily. The most important factor for a relapse in alcohol drinking was shown to be if the patient already had had one relapse during the treatment. Other important clinical factors were the levels of irritation and autonomic disturbances. None of the variables measuring mood shifts was significant. Family conditions during childhood were the most important background variables. The predictions turned out to have a rather high specificity, but the sensitivity was lower. Half of the relapses were not predicted by an increased probability for relapse. Self-reported relapses were predictable from preceding interviews and were also less frequent compared to those detected objectively by the biochemical markers.