630 patients with haematemesis and melaena were randomly allocated to treatment by a constant intravenous infusion of either somatostatin or an apparently identical placebo in a double-blind controlled trial. Rebleeding was less common in treated patients (70 episodes in 315 individuals compared with 89 episodes in 315 controls) but the difference was not significant. Operation rates were virtually identical (35 treated patients and 34 controls), while there were slightly more deaths in the treated group than in the controls (31 and 25, respectively). These results are in clear disagreement with those of other smaller series. Though it is not possible to be completely sure that treatment is not useful in some individuals, earlier claims of marked benefit seem unlikely to be justified.