Fifty-two patients with severe cirrhosis (Child Class C) and variceal hemorrhage requiring six or more units of blood were randomly assigned to either sclerotherapy or portacaval shunt. Of 38 pretreatment characteristics, only the frequency of active alcoholism differed significantly between the groups. During the initial hospitalization, the patients in the shunt group required significantly more blood (21.5 +/- 3.1 units) than did those in the sclerotherapy group (12.3 +/- 1.3 units), although the latter had significantly more rebleeding during hospitalization after the procedure (14 of 28 vs. 5 of 24 patients). There was no difference in short-term survival, with 13 patients in the sclerotherapy group discharged alive, as compared with 10 patients in the shunt group. Patients were followed for a mean of 263 days after the initial discharge (range, 8 to 1117). The sclerotherapy group required significantly more days of hospitalization for rebleeding, but we failed to demonstrate any significant difference in long-term survival between the sclerotherapy and shunt groups. Total health-care costs per patient were significantly higher for the shunt group (+23,957 +/- +3,111) than for the sclerotherapy group (+15,364 +/- +2,220). We conclude that sclerotherapy is less costly than portacaval shunt and as effective for the treatment of esophageal varices associated with severe cirrhosis.