Endoscopic injection therapy was performed in a consecutive series of 233 patients admitted for a bleeding peptic ulcer with active arterial hemorrhage or a nonbleeding visible vessel disclosed at emergency endoscopy. Further bleeding occurred in 57 cases (24.5%). The present study was conducted to evaluate whether any clinical or endoscopic features could identify the patients at high risk of therapeutic failure. Multiple logistic regression analysis showed that failure was significantly related to: (1) the ulcer location on the posterior wall (P = 0.004) or superior wall (P = 0.003) of the duodenal bulb, (2) the ulcer size (P = 0.011), and (3) the existence of associated diseases (P = 0.012). The validity of the prediction rule based on these factors was evaluated by receiver-operating characteristic curves and was confirmed and prospectively validated in an independent sample of 81 patients with a bleeding peptic ulcer treated by endoscopic injection. We conclude that once the initial control of bleeding has been achieved by injection therapy, the present prediction rule can be used to identify candidates for alternative treatment.