The aim of this study was to establish the relative importance of risk factors for mortality after acute upper gastrointestinal haemorrhage, and to formulate a simple numerical scoring system that categorizes patients by risk. A prospective, unselected, multicentre, population based study was undertaken using standardised questionnaires in two phases one year apart. A total of 4185 cases of acute upper gastrointestinal haemorrhage over the age of 16 identified over a four month period in 1993 and 1625 cases identified subsequently over a three month period in 1994 were included in the study. It was found that age, shock, comorbidity, diagnosis, major stigmata of recent haemorrhage, and rebleeding are all independent predictors of mortality when assessed using multiple logistic regression. A numerical score using these parameters has been developed that closely follows the predictions generated by logistical regression equations. Haemoglobin, sex, presentation (other than shock), and drug therapy (non-steroidal anti-inflammatory drugs and anticoagulants) are not represented in the final model. When tested for general applicability in a second population, the scoring system was found to reproducibly predict mortality in each risk category. In conclusion, a simple numerical score can be used to categorize patients presenting with acute upper gastrointestinal haemorrhage by risk of death. This score can be used to determine case mix when comparing outcomes in audit and research and to calculate risk standardised mortality. In addition, this risk score can identify 15% of all cases with acute upper gastrointestinal haemorrhage at the time of presentation and 26% of cases after endoscopy who are at low risk of rebleeding and negligible risk of death and who might therefore be considered for early discharge or outpatient treatment with consequent resource savings.