OBJECTIVE The Glasgow-Blatchford score (GBS) has been validated to select severe patients with non-variceal upper gastrointestinal hemorrhage (UGIH). The aim was to compare the yield of the triage based on the GBS with an endoscopist' decision to perform an urgent upper gastrointestinal endoscopy (UGIE) in newly admitted patients and inpatients with UGIH in the setting of an endoscopy on-duty service in 13 tertiary care centers. METHODS During a 6-month period, GBS and patient data were collected for all patients with non-variceal UGIH for whom an UGIE was requested in emergency. If patients experienced severe endoscopic lesion, surgery or death, they were categorized as patients who had been at need for urgent UGIE. RESULTS The 102 UGIH patients included (mean age 62, men 73%) had a median GBS of 12 (range 0-21), significantly lower for new patients compared with inpatients (11, range 0-21 vs. 14, range 2-21, respectively, p = 0.001). If triage for urgent UGIE had followed the GBS, no more patients would have had an urgent UGIE compared with what endoscopists performed (99/102 (97%) vs. 92/102 (90%), respectively, p = 0.09). Sensitivity for the detection of patients who needed an UGIE was no different with the GBS than endoscopists (98% vs. 98%, respectively, p = 0.10) and both showed insufficient specificity (4% and 19%, respectively). CONCLUSIONS The GBS does not detect more patients at need for urgent UGIE than on-duty endoscopists. Both methods lead to numerous unjustified UGIEs. A score that would equally help endoscopists in their decision to intervene urgently is still warranted.