One of the most promising nonoperative techniques for control of variceal hemorrhage is sclerosis via the fiberoptic endoscope. Many questions remain, however, about sclerosing agents, guidelines for effective use, and limitations of endoscopic techniques. A reproducible large animal model of esophageal varices would facilitate the critical evaluation of techniques for variceal hemostasis or sclerosis. Our purpose was to develop a large animal model of esophageal varices. Studies in pigs and dogs are described which led to the development of a reproducible canine model of esophageal varices. For the final model, mongrel dogs had laparotomy, side-to-side portacaval shunt, inferior vena cava ligation, placement of an ameroid constrictor around the portal vein, and liver biopsy. The mean (+/- SE) pre- and postshunt portal pressure increased significantly from 12 +/- 0.4 to 23 +/- 1 cm saline. Weekly endoscopies were performed to grade the varix size. Two-thirds of animals developed medium or large sized esophageal varices after the first operation. Three to six weeks later, a second laparotomy with complete ligation of the portal vein and liver biopsy were performed in animals with varices (one-third of the animals). All dogs developed esophageal varices and abdominal wall collateral veins of variable size 3-6 wk after the first operation. After the second operation, the varices became larger. Shunting of blood through esophageal varices via splenic and gastric veins was demonstrated by angiography. Sequential liver biopsies were normal. There was no morbidity or mortality. Ascites, encephalopathy, or spontaneous variceal bleeding did not occur. We have documented the lack of size change and the persistence of medium to large esophageal varices and abdominal collateral veins in all animals followed for more than 6 mo. Variceal bleeding could be induced by venipuncture for testing endoscopic hemostatic and sclerosis methods. We suggest other potential uses of this reproducible canine model of esophageal varices.