The aim of this study was to prospectively evaluate the usefulness of the Italian Liver Cirrhosis Project (ILCP) classification of esophageal varices, together with the gastric features of portal hypertension, in predicting the first upper gastrointestinal bleeding in cirrhosis. The efficiency of these endoscopic parameters in predicting bleeding was also compared with the efficiency of the North Italian Endoscopic Club (NIEC) index. Three hundred forty‐four cirrhotic patients with esophageal varices but without any previous bleeding were enrolled in the study. The following endoscopic parameters of esophageal varices were recorded: location, size, occupancy, blue tone, and red color signs. Gastric varices were graded as absent or present, while congestive gastropathy was considered as absent, mild to moderate, or severe. All patients were followed until the first upper gastrointestinal bleeding and/or death, or for at least 24 months. No patient received any treatment to prevent bleeding. Sixty‐five patients bled from the upper gastrointestinal tract during the study. Univariate analysis showed that all endoscopic parameters were predictors of bleeding. According to multivariate analysis (Cox's model), size, gastric varices and congestive gastropathy were the only independent predictors of bleeding, and the following prognostic index (PI) was developed: PI = (size × 0.0395) + (congestive gastropathy × 0.878) + (gastric varices × 0.705). This index, which was validated using a split‐sample technique, and which appears to be superior to the NIEC index in predicting bleeding, may be useful in decision making for primary prophylaxis.