Aim Acute gastrointestinal bleeding is a potentially life‐threatening condition that requires rapid assessment and dynamic management. Several scoring systems are used to predict mortality and rebleeding in such cases. The aim of the present study was to compare three scoring systems for predicting short‐term mortality, rebleeding, duration of hospitalization and the need for blood transfusion in elderly patients with upper gastrointestinal bleeding. Methods The present study included 335 elderly patients with upper gastrointestinal bleeding. Pre‐ and post‐endoscopic Rockall, Glasgow–Blatchford and AIMS65 scores were calculated. The ability of these scores to predict rebleeding, mortality, duration of hospitalization and the need for blood transfusion was determined. Results Pre‐ (4.5) and post‐endoscopic (7.5) Rockall scores were superior to the Glasgow–Blatchford (12.5) score for predicting mortality (P = 0.006 and P = 0.015). Likewise, pre‐ (4.5) and post‐endoscopic Rockall scores were superior to the respective Glasgow–Blatchford scores for predicting rebleeding (P = 0.013 and P = 0.03). There was an association between duration of hospitalization and mortality; as the duration of hospitalization increased the mortality rate increased. In all, 94% of patients hospitalized for a mean of 5 days were alive versus 56.1% of those hospitalized for 20 days, and 20.2% of those hospitalized for 40 days. Conclusions In elderly patients with upper gastrointestinal bleeding, the Rockall score is clinically more useful for predicting mortality and rebleeding than the Glasgow–Blatchford and AIMS65 scores; however, for predicting duration of hospitalization and the need for blood transfusion, the Glasgow–Blatchford score is superior to the Rockall and AIMS65 scores. Geriatr Gerontol Int 2017; 17: 575–583.