In order to determine immediate criteria of prognosis for patients with portal hypertension hospitalized for digestive hemorrhage, in an intensive care unit, 18 variables were recorded during the 24 hours following admission in 65 patients. Data related to death were age, ascites, hepatic encephalopathy, shock, active hemorrhage, acute pneumonia, decrease in prothrombin time, use of esophageal balloon tamponade, use of mechanical ventilation, number of red blood cell units transfused. Discriminant analysis yielded a linear combination of 4 variables which best separated survivors from non survivors with the following equation: F = 0.330 X hepatic encephalopathy + 0.433 X shock + 0.226 X active hemorrhage + 0.0097 X age - 0.396. The threshold decision of the hemorrhage prognosis index (HPI) was F = 0.57; 80 p 100 of all patients were correctly classified. In order to be validated, HPI was compared with a general (SAPS) and specific (Pugh's classification) scoring system, in a prospective study of 57 episodes of digestive hemorrhage. In this study, sensitivity was better with HPI than with SAPS (0.70 versus 0.45), specificity was higher with HPI than with Pugh's classification (0.86 versus 0.70). Percentage of correctly classified patients was higher using HPI (81 p. 100) than SAPS (77 p. 100) and Pugh's classification (68 p. 100). We suggest that the HPI, determined with 4 easily defined and recorded variables should be used prospectively to compare efficacy of different treatments.