Hemorrhage from esophageal varices in cirrhotics is a frequent event with high mortality in spite of therapy. Preventive sclerotherapy seems to be beneficial only if the patient's bleeding risk is higher than 40 to 50% a year. A series of 320 patients with esophageal varices without previous bleeding was studied prospectively; the varices were classified according to three widely used endoscopic classifications. During follow-up (6 to 36 months, average 14 months), hemorrhage occurred in 49 patients (15.3%) of whom 30 (61.2%) bled from varices (8.2 and 11.0% at 12 and 24 months, respectively). At the same time intervals, mortality of the entire population studied was 18.0 and 23.8%, respectively, of which one third was directly due to hemorrhage. With all three classifications, the higher the degree of bleeding risk, the greater the actual percentage of hemorrhages recorded; however, it never reached 40% a year. In predicting the bleeding event, Dagradi's classification proved more sensitive than JRSPH or NIEC, but the latter classifications were more specific and assessed a higher predictive value for a positive test. Endoscopic observation probably needs integration with other methods if a reliable bleeding prediction is to be made.