The objective of this study was to assess the prognostic validity of Child-Turcotte classification with regard to short-term (1-year) survival. The Child-Turcotte classification, as modified by Pugh et al., was recorded on admission in 177 cirrhotic patients. The variables that comprise the Pugh modification are ascites, encephalopathy, serum albumin, serum bilirubin and prothrombin time. Using multiple logistic regression, we evaluated the contribution of different models to the likelihood of survival, defining different ways to use the Pugh score. The Pugh score categorized in three strata (5 to 6, 7 to 9 and 10 to 15) captured less variance in the survival than the Pugh score counted from 5 to 15. This, in turn, captured less variance than a model in which the parameters of the Pugh score were analyzed according to their original units. The prediction rule based on the last model was tested in another sample of cirrhotics. The "original unit" model was studied in both training and testing samples, using receiver-operating characteristic curves to evaluate its clinical validity (sensitivity and specificity). The prediction rule based on the "original units" Pugh score allowed for a good discrimination of patients who lived and those who died. (At the point of maximum discrimination, sensitivity and specificity reached a mean of 80%). Validity of the prediction rule was confirmed by the testing sample. The qualities of simplicity, availability, low cost and good discriminating power for a life or death outcome make the Pugh score a very useful method to estimate prognosis in patients with cirrhosis.