Twenty-one patients with duodenal ulcer were treated with cimetidine. After 2 weeks they were submitted to an insulin test and the results were compared with a range of secretion previously established in untreated duodenal ulcer patients (1). The secretion of 13 patients fell within this range and so they were predicted to have a high risk of relapse. The patients whose secretion was below the range had a predicted low risk of relapse. After 4 weeks' treatment repeat endoscopy showed that 4 ulcers had failed to heal. The secretion of 3 of the patients was within the range of the high risk group. The remaining 17 patients were followed up (range 20-23 months): 9 relapsed and 8 remained symptom-free. Eight of the 9 were in the high risk group and 6 of the 8 symptom-free patients were in the low risk group. In all, 11 out of 13 were correctly predicted in the high risk group and 6 out of 8 in the low risk group. The performance of the insulin test in predicting liability to relapse following cimetidine treatment was significant (P = 0.0225).