Variceal haemorrhage in cirrhotic patients carries a high early mortality even when balloon tamponade or emergency sclerotherapy are applied. The aim of this study to identify patients dying within six weeks of their first variceal haemorrhage. One hundred and twenty one patients with parenchymal cirrhosis presenting with the first variceal bleeding episode between June 1983 and December 1988 were studied. Nineteen patients were excluded for various reasons. Emergency sclerotherapy was carried out in cases of active bleeding or where there were endoscopic signs of recent bleeding, and then regularly repeated afterwards. Of the 24 variables studied and included in a multivariate analysis using a logistic regression model, three had an independent prognostic value: encephalopathy, prothrombin time, and the number of blood units transfused within the 72 hours of time zero. The subsequent regression equation was able to predict 89% of the patients who will die and 97% of the patients who will still be alive six weeks after their first variceal haemorrhage treated by sclerotherapy. Pugh score was less discriminatory than these last three variables in terms of accuracy of adjustment, goodness of fit to the model, receiver operating characteristic curves, and percentage correct prediction. To measure the accuracy of the prediction rule, our model was applied to another series of 28 cirrhotic patients admitted with their first variceal bleeding during the next period (January 1989 to May 1990). Death and survival were correctly predicted in respectively 82% and 94% of the cases. The use of this score is recommended for the selection of patients with high early mortality after variceal bleeding despite sclerotherapy, and for the design of new therapeutic trials.