We prospectively and randomly compared heat probe and ethanol injection in 80 patients with major nonvariceal upper gastrointestinal hemorrhage who were bleeding actively or had endoscopic stigmata associated with a high risk for rebleeding. We also attempted to predict which patients would rebleed within 72 h after successful endoscopic therapy, using a three-component scoring system. Heat probe and ethanol injection proved to be similar in efficacy and safety. Active bleeding was controlled with equal success with heat probe and ethanol injection (92% vs. 82%), and there was no difference in the rebleeding rate (11% vs. 13%). The scoring system was useful in predicting which patients would rebleed. Significant differences were seen in the mean values of all three scores, and specific cut-offs in the pre-endoscopy and post-endoscopy scores predicted patients who rebled. High likelihood ratios and post-test probabilities for rebleeding were found for the number and severity of concurrent illnesses, but not for endoscopic stigmata, implying that the excess risk associated with stigmata is eliminated after effective endoscopic therapy, and clinical factors become the primary determinants of rebleeding.