Between January 1983 and December 1985, 305 patients were admitted to our hospital because of bleeding duodenal ulcer. A subgroup of 69 patients aged 50 or above in whom emergency endoscopy showed non-arterial bleeding or signs of recent haemorrhage without a visible vessel entered a prospective therapeutic trial. The patients were randomized to receive either (1) early surgery, implying immediate operation, or (2) expectant management, with surgery reserved only for patients with further haemorrhage. The two groups were homogeneous with respect to age, sex, prior ingestion of ulcerogenic drugs, mode of bleeding, admission haematocrit, number with hypovolaemic shock and number with active bleeding on initial endoscopy. Overall mortality was 8.6 per cent. Mortality in patients submitted to early surgery was five times higher than that in those allocated to expectant therapy (14.7 per cent versus 2.9 per cent; risk ratio 5.07). The results suggest that expectant management is advisable in patients with bleeding duodenal ulcer not bleeding massively and in whom endoscopy does not disclose spurting arterial bleeding or a visible vessel.