Two types of surgical therapy of bleeding esophageal varices were evaluated in 48 patients by a randomized controlled trial: 24 were randomized for a total shunt and 24 for the selective shunt. In two of the latter, a total shunt had to be performed for technical reasons. The fatality rates (six in the 24 total, and six in 22 selective [performed], and seven in 24 selective [randomized]), the frequency of shunt occlusion (two in each group), and of recurrent gastronintestinal bleeding (three in each group) were similar. Encephalopathy developed more often after a total shunt -- 10 of 24, or one per 58 patient-months -- than after selective (performed) -- one of 22, or one per 593 patient-months (P less than 0.005). Total shunts consistently diverted the hepatopetal mesenteric-portal flow from the liver. Deterioration of hepatic function (maximum rate of urea synthesis) was greater after total than selective shunt (P less than 0.05).