The natural history of gastrointestinal bleeding in cirrhosis has been studied using prospectively collected data of 532 patients included in a randomized clinical trial with a regular follow-up of up to 12 yr. Of the total 199 patients who experienced gastrointestinal bleeding, 95 (48%) bled from esophageal or gastric varices, 67 (34%) bled from peptic ulcer or gastritis, and 37 (18%) had either insufficient evidence of the source (33) or mixed sources (4). In the total group of patients the cumulative percentage of patients in whom varices had been demonstrated of patients in whom varices had been demonstrated by radiography increased from 12 to 90 in 10 yr, while that of bleeding from varices increased from 7 to 40. In 104 patients who bled for the first time during the trial period (trial bleeding patients) the median number of bleeding episodes was one (range 1-8). In these patients the fatality from bleeding from varices was 82%. The risk of rebleeding from varices was 81%, and 4 yr after the first bleeding the cumulative survival had decreased to less than 10%. Rebleeding was significantly less frequent and survival significantly higher in patients bleeding from sources other than varices. Prednisone reduced the occurrence rate of varices, bleeding from varices, and death from bleeding varices in nonalcoholic females without ascites, 40% of whom fulfilled the histologic criteria of chronic active hepatitis. Prednisone significantly increased the occurrence rate of varices inpatient with ascites and of bleeding from varices in alcoholic patients. Prednisone significantly increased the occurrence rate of peptic ulcer in males and in patients without chronic active hepatitis.