This study predicts the outcomes of 40 men and 20 women who attended a controlled drinking programme within a general hospital. Drinking levels over a 6-month follow-up were significantly predicted by consumption during the programme, by the severity of alcohol problems in the past and by a new measure of self-efficacy level that was administered at post-treatment. The self-efficacy prediction was still significant even after all of the other variables were forced to enter a prediction equation. Results are discussed in terms of a social-cognitive theory of sustained response to treatment.