Objective: the aim of this study was to determine the usefulness of Rockall score in predicting outcomes of 30 days rebleeding, mortality and need for surgical intervention of bleeding gastric and duodenal ulcers. Methods: this is a retrospective cohort study of all the emergency endoscopies performed in Hospital sultan Ismail from January 2009 to October 2014 for indications of upper gastrointestinal bleeding (UGIb). Data was extracted from hospital's electronic database and only non-variceal bleeds were included. Rockall score was calculated and outcomes of 30 days rebleeding, mortality and need for surgery was recorded. For each outcome, calibration was done using the Goodness-of-fit tests and discriminative ability was reflected by area under the receiver operating characteristic curve (AUROc). Results: A total of 1323 patients were included with a male preponderance of 64%. the overall rates of rebleeding were 11.2%, mortality rate of 8.7% and need for surgery was 2%. Low AUROc values for rebleeding (0.63), mortality (0.58) and surgery (0.67) showed poor discriminative ability of Rockall score. the Goodness-of-fit test also revealed that the scoring system was poorly calibrated in outcomes of rebleeding (p <0.001), mortality (p = 0.001) and surgery (p = 0.038) with p-value <0.05. Patients with high risk (scores ≥8) displayed highest rebleeding and mortality rates of 20% respectively in comparison to the moderate (score 3-7) and low (score ≤2) risk groups. conclusion: Rockall score has a poor discriminative ability and is poorly calibrated for rebleeding, mortality and need for surgery in upper gastrointestinal bleeding. However, it is the best tool we have now to stratify patients into risk groups.