OBJECTIVE: Recent guidelines recommend that all cirrhotics undergo screening upper endoscopy to identify those patients at risk for bleeding from varices. However, this practice may not be cost effective as large esophageal varices are seen only in 9-36% of these patients. The aim of this study was to determine whether clinical variables were predictive of the presence of large esophageal varices. METHODS: This is a retrospective analysis of cirrhotics who had a screening upper endoscopy during an evaluation for liver transplantation at three different centers and who had not previously bled from varices. A multivariate model was derived on the combined cohort using logistic regression. Three hundred forty-six patients were eligible for the study. RESULTS: The prevalence of large esophageal varices was 20%. On multivariate analysis, splenomegaly detected by computed tomographic scan (odds ratio: 4.3; 95% confidence interval: 1.6-11.5) or by physical examination (odds ratio: 2.0; 95% confidence interval: 1.1-3.8), and low plate- let count were independent predictors of large esophageal varices. On the basis of these variables, cirrhotics were stratified into high- and low-risk groups for the presence of large esophageal varices. Patients with a platelet count of ≥88,000/mm3 (median value) and no splenomegaly by physical examination had a risk of large esophageal varices of 7.2%. Those with splenomegaly or platelet count < 88,000/mm3 had a risk of large esophageal varices of 28% (p < 0.0001). CONCLUSIONS: Our data show that clinical predictors could be used to stratify cirrhotic patients for the risk of large esophageal varices and such stratification could be used to improve the cost effectiveness of screening endoscopy.