Histamine H2 antagonists are widely used in treating patients with hematemesis and melena, despite the lack of reliable evidence of benefit from any of the randomized trials, considered separately. Examination of the data from all 27 available randomized trials, in which over 2500 patients were entered, suggests that treatment may reduce the rates of rebleeding, surgery, and death by about 10, 20, and 30 per cent, respectively, although these results were only marginally significant for surgery and death. Any benefit appeared to be confined to patients with bleeding gastric ulcers, but since this subgroup analysis was prompted by preliminary examination of the data in some of the individual trials reviewed here, it should be treated with particular caution. The implications of this overview are that treatment with histamine H2 antagonists appears to be moderately promising, but its effects on important end points, such as death, still need to be assessed reliably. Prevention of "only" about 20 per cent of all deaths could well be worthwhile, for the condition is common, and the treatment widely practicable. To detect such a moderate effect reliably, however, might require the randomization of 10,000 patients (or more), which would be possible only in an extremely simple multicenter collaborative trial.