Our aims were to develop a noninvasive predictive tool to identify cirrhotic patients with esophageal varices and to evaluate whether portal Doppler ultrasonographic parameters may improve the value of other predictors. One hundred forty-three consecutive compensated cirrhotic patients underwent upper gastrointestinal endoscopy. Fourteen clinical, biochemical, ultrasonographic, and Doppler ultrasonographic parameters of each patient were also recorded. Esophageal varices were detected in 63 of the 143 patients examined (44%; 95% confidence interval [CI] 36.2-52.6). Medium and large esophageal varices were observed in 28 subjects (44%; 95% CI 31.4-58.4). Using stepwise logistic regression, presence of esophageal varices was independently predicted by prothrombin activity less than 70% (odds ratio [OR]: 5.83; 95% CI: 2.6-12.8), ultrasonographic portal vein diameter greater than 13 mm (OR: 2.92; 95% CI: 1.3-6.4), and platelet count less than 100 x 10(9)/L (OR: 2.83; 95% CI: 1.27-6.28). Variables included in the model were used to generate a simple incremental rule to evaluate each individual patient. The discriminating ability of the prediction rule was relevant (area under the curve: 0.80) and did not change by replacing ultrasonographic portal vein diameter with congestion index of portal vein. We concluded that compensated cirrhotic patients should be screened by upper gastrointestinal endoscopy when prothrombin activity less than 70%, platelet count less than 100 x 10(9)/L, and ultrasonographic portal vein diameter greater than 13 mm are observed, whereas those without any of these predictors should not undergo endoscopy. The contribution provided by portal Doppler ultrasonographic parameters does not appear of practical utility.