Endoscopic haemostasis by injection of adrenaline was attempted in 135 consecutive patients with active upper gastrointestinal bleeding. Initial haemostasis was obtained in 127 patients following injection of 5-15 ml 1:10,000 adrenaline; eight patients in whom haemostasis was not achieved underwent immediate laparotomy. There was further haemorrhage in 25 patients, which was successfully treated by further injection of adrenaline in ten. Fifteen patients had major rebleeding requiring emergency surgery. Stepwise logistic regression analysis identified three factors that, taken together, were highly predictive of the need for surgery: pulse rate on admission, the position of the ulcer and whether the patient was obese. A scoring system was derived from the logistic analysis equation that was found to predict correctly the need for emergency surgery in 84 per cent of patients. In patients with a high probability of rebleeding surgery should be considered after initial endoscopic haemostasis and stabilization. In the majority of patients endoscopic treatment alone is sufficient for permanent haemostasis.