Background Upper gastrointestinal bleeding (UGIB) is a medical emergency with an approximate mortality of 10%, which results in a high hospitalisation rate. The Glasgow-Blatchford score (GBS) is recommended to identify low-risk patients who can be discharged from the emergency department (ED). A modified GBS (mGBS) and CANUKA score have recently been proposed but have not been well studied. The aim of this study was to assess whether the use of GBS, mGBS or CANUKA score could identify patients at low risk of death or need for intervention. Methods A single-centre retrospective study was performed including patients with suspected UGIB visiting the ED of Saint-Antoine hospital (Paris, France) from January 2016 to December 2018. Demographic and medical data needed to calculate GBS and CANUKA were collected, as well as outcomes data. Need for intervention was defined as the need for blood transfusion, endoscopic haemostasis or rebleeding within 7 days. In-hospital mortality was also collected. Sensitivity, specificity and predictive values were measured for the score thresholds of interest. Results A total of 386 patients were included. Median age was 60 years (38–78), 65.3% (n=252) were male and 60% (n=233) were hospitalised. A GBS≤1, mGBS=0 and CANUKA≤2 categorised 24.9%, 18.2% and 18.9% of patients as low risk, respectively. There was a need for intervention in 2.2%, 4.6% and 0% of those patients categorised as low risk by GBS, mGBS and CANUKA, respectively. No deaths occurred in the patients identified as low risk, regardless of the score used. All scores had a high sensitivity and negative predictive value. Conclusions In patients with UGIB, the use of a GBS≤1 or CANUKA score ≤2 appears to be safe for identifying patients at low risk of death or need for intervention.