Background/Aims: This prospective study investigated the performance of pre-endoscopy and the complete Rockall scores in predicting the occurrence of adverse outcomes and the need for endoscopic or surgical intervention in patients with nonvariceal upper gastrointestinal bleeding. Methods: All 656 consecutive patients who underwent endoscopy due to nonvariceal upper gastrointestinal bleeding between 2007 and 2011 were included. Receiver operating characteristic (ROC) curves were plotted for the outcomes of therapeutic intervention, rebleeding and death. The discriminative accuracy of the risk scores was assessed by the area under the ROC curve. Results: Endoscopic treatment was performed in 55.2% of the patients. Rebleeding and mortality rates were 7.6 and 3.8%, respectively. The pre-endoscopy Rockall scores showed unsatisfactory accuracy in predicting the need for intervention, rebleeding or death, as shown by the respective area under the ROC curve values of 0.52, 0.52 and 0.65. The accuracy of the complete Rockall score in predicting rebleeding was poor (area under ROC: 0.52), but it was higher for mortality (area under ROC: 0.69). Conclusions: The pre-endoscopy Rockall score was not useful for predicting the need for therapeutic intervention or adverse outcomes. The complete Rockall score showed an acceptable performance in predicting mortality, but was unable to predict rebleeding.