Over a four-year period, 585 patients were hospitalized for massive upper gastrointestinal bleeding. Endoscopy diagnosed the cause of bleeding in 80% of 200 patients so studied. Selective angiography localized the bleeding site in 12 of 20 patients, and infusion of vasopressor stopped hemorrhage in six. Barium studies was 90% accurate in diagnosing ulcer disease but failed to detect gastritis. One hundred thirty (22%) patients were operated upon for medically uncontrolled bleeding. The proportion of patients requiring surgery fell from 33% in year one to 13% in year four. Benign ulcer disease caused bleeding in 51% of surgical patients, while gastritis was found in 20%, esophageal varices in 15% and stress ulcer in 8%. Overall operative mortality was 29%. Among 38 duodenal ulcer patients, mortality was 18%. Vagotomy and pyloroplasty were more effective than resection in this group. Resection for distal gastric ulcers in 22 patients resulted in a mortality of 14% and no rebleeding. While V&P controlled bleeding in 12 alcoholics with gastritis, five (42%) died postoperatively. Mortality among 20 patients with esophageal varices was 35%, although all five survived who had porto-caval shunts. Eight of 10 patients operated upon for stress ulcer bleeding died. Postoperative rebleeding occurred in 14 patients, eight of whom were again operated upon. In all but one a new lesion was found to be responsible for hemorrhage. Increasing use of gastroscopy and selective angiography can be expected to improve diagnostic capabilities in patients with upper gastrointestinal bleeding. Infusing vasopressor into selected arteries should reduce the need for surgical control of gastritis, variceal and stress ulcer bleeding, conditions poorly managed by current operative techniques.