Introduction. Management of upper gastrointestinal bleeding (UGIB) is of great importance. In this way, we aimed to evaluate the performance of three well known scoring systems of AIMS65, Glasgow-Blatchford Score (GBS) and Full Rockall Score (FRS) in predicting adverse outcomes in patients with UGIB as well as their ability in identifying low risk patients for outpatient management. We also aimed to assess whether changing albumin cutoff in AIMS65 and addition of albumin to GBS add predictive value to these scores. Methods. This was a retrospective study on adult patients who were admitted to Razi hospital (Rasht, Iran) with diagnosis of upper gastrointestinal bleeding between March 21, 2013 and March 21, 2017. Patients who didn’t undergo endoscopy or had incomplete medical data were excluded. Initially, we calculated three score systems of AIMS65, GBS and FRS for each patient by using initial Vital signs and lab data. Secondary, we modified AIMS65 and GBS by changing albumin threshold from <3.5 to <3.0 in AIMS65 and addition of albumin to GBS, respectively. Primary outcomes were defined as in hospital mortality, 30-day rebleeding, need for blood transfusion and endoscopic therapy. Secondary outcome was defined as composition of primary outcomes excluding need for blood transfusion. We used AUROC to assess predictive accuracy of risk scores in primary and secondary outcomes. For albumin-GBS model, the AUROC was only calculated for predicting mortality and secondary outcome. The negative predictive value for AIMS65, GBS and modified AIMS65 was then calculated. Result. Of 563 patients, 3% died in hospital, 69.4% needed blood transfusion, 13.1% needed endoscopic therapy and 3% had 30-day rebleeding. The leading cause of UGIB was erosive disease. In predicting composite of adverse outcomes all scores had statistically significant accuracy with highest AUROC for albumin-GBS. However, in predicting in hospital mortality, only albumin-GBS, modified AIMS65 and AIMS65 had acceptable accuracy. Interestingly, albumin, alone, had higher predictive accuracy than other original risk scores. None of the four scores could predict 30-day rebleeding accurately; on the contrary, their accuracy in predicting need for blood transfusion was high enough. The negative predictive value for GBS was 96.6% in score of ≤2 and 85.7% and 90.2% in score of zero in AIMS65 and modified AIMS65, respectively. Conclusion. Neither of risk scores was highly accurate as a prognostic factor in our population; however, modified AIMS65 and albumin-GBS may be optimal choice in evaluating risk of mortality and general assessment. In identifying patient for safe discharge, GBS ≤ 2 seemed to be advisable choice.