A prospective study was conducted to collect data on clinical and endoscopic diagnosis, associated factors and outcome of 112 consecutive patients with acute upper gastrointestinal haemorrhage admitted to the Dunedin public hospitals over an 18 month period. The mean interval between admission and endoscopy was 20.2 hours. There was a poor correlation between the provisional clinical diagnosis and the endoscopic diagnosis. The causes of bleeding were demonstrated at endoscopy in 87.5% of patients. A history of salicylate, non-steroidal anti-inflammatory drug or significant alcohol ingestion was present in about two-thirds of patients with mucosal abnormalities. The incidence of continued or repeated bleeding was 16%; peptic ulcers accounted for half of these patients. The presence of active bleeding or a visible vessel or blood clot on the ulcer surface indicated a 33% chance of rebleeding; none of the peptic ulcer patients without these signs rebled. Surgery was performed in 11.6% of patients, predominantly for peptic ulcer. The overall mortality was 8% (reducing to 5.4% if patients dying in hepatic failure are excluded), most of the deaths occurring in older patients with complicating medical conditions, and not from uncontrolled haemorrhage.