Evidence for the effectiveness of existing treatments of patients with eating disorders is weak. Here we describe and evaluate a method of treatment in a randomized controlled trial. Sixteen patients, randomly selected out of a group composed of 19 patients with anorexia nervosa and 13 with bulimia nervosa, were trained to eat and recognize satiety by using computer support. They rested in a warm room after eating, and their physical activity was restricted. The patients in the control group (n = 16) received no treatment. Remission was defined by normal body weight (anorexia), cessation of binge eating and purging (bulimia), a normal psychiatric profile, normal laboratory test values, normal eating behavior, and resumption of social activities. Fourteen patients went into remission after a median of 14.4 months (range 4.9–26.5) of treatment, but only one patient went into remission while waiting for treatment (P = 0.0057). Relapse is considered a major problem in patients who have been treated to remission. We therefore report results on a total of 168 patients who have entered our treatment program. The estimated rate of remission was 75%, and estimated time to remission was 14.7 months (quartile range 9.6 ≥ 32). Six patients (7%) of 83 who were treated to remission relapsed, but the others (93%) have remained in remission for 12 months (quartile range 6–36). Because the risk of relapse is maximal in the first year after remission, we suggest that most patients treated with this method recover.