A reliable tool to predict endoscopic stigma of recent hemorrhage (SRH) and need for ET based on clinical information available at initial evaluation of patients with UGIB would be useful for triaging high-risk patients for urgent endoscopy. Aim: To evaluate role of an ANN based predictive model in predicting SRH and need for ET in patients with UGIB. Methods: Clinical data on all patients admitted with UGIB during a 15 month period starting January 1998 were prospectively collected. A multi-layered perceptron ANN was constructed using clinical variables available at initial evaluation. The ANN was trained by back-propagation using data from initial 9 months (Group I) and then validated using data from the last 6 months (Group II). The ANN was blinded to the endoscopic finding of patients from Group II. For comparison, a stepwise logistic regression model was constructed using the available clinical input variables. Also, two validated scoring systems viz. Baylor pre-endoscopic score (Am J Gastroenterol 1993) and Decision rule by Corley et al ( Am J Gastroenterol 1998) were used to classify patients into high and low risk groups. Presence of SRH and need for ET were the two outcome parameters used to compare the different predictive models. Results: 327 patients (Mean age 66.2 years, 169 males) were available for analysis. The clinical features were similar in Group I (n = 208) and Group II (n = 119). SRH was present in 110 (33.6%) patients and ET was performed in 105 (32.1%). ANN performed better than other models (Table). Area under the receiver operating curve for ANN were 0.89 and 0.93 for presence of SRH and ET, respectively. Conclusion: ANN performed extremely well in predicting presence of SRH and need for ET from information available at initial presentation. ANN may be a useful tool in triaging patients with UGIB for urgent endoscopy.