A prospective and randomized trial involving 104 patients was performed to assess whether second-look endoscopy could improve the efficacy of injection therapy for bleeding ulcers. The inclusion criteria were the presence of active arterial bleeding or a non-bleeding visible vessel at emergency endoscopy. All the patients received emergency injection of 1:10,000 adrenaline and were subsequently randomized (52 patients in each group) according to whether or not they would receive a second elective endoscopy within the first 24 hours with repeated injection if a visible vessel was still identified. Both groups were well matched for clinical and endoscopic data. A tendency towards better results was noted in the group that received a second-look endoscopy; the two groups were compared in regard to further bleeding (21% versus 29%, 95% confidence interval of the difference = -24.3 to 8.5), need for emergency surgery (8% versus 15%, 95% confidence interval of the difference = -19.9 to 4.5), transfusion requirements (1.7 +/- 1.9 versus 2.5 +/- 2.5 units, 95% confidence interval of the difference = -1.6 to 0.07), length of hospital stay (9.3 +/- 8.6 versus 11.8 +/- 10.8 days, 95% confidence interval of the difference = -6.2 to 1.4), and mortality rate (2% versus 4%). Although these trends did not achieve statistical significance, a type II error cannot be ruled out. However, according to our results, several hundred patients would be required to demonstrate statistically these relatively small differences.(ABSTRACT TRUNCATED AT 250 WORDS)