Portal hypertension with varices developed in 18/675 patients with chronic myeloid leukaemia (CML) in a randomized trial comparing busulphan with busulphan and thioguanine. All 18 had received the drug combination and none busulphan alone (P less than 0.0001). Ascites was also seen significantly more often in the combination arm (P less than 0.05). These results strongly suggest that the addition of thioguanine was responsible for the development of portal hypertension. The histological features were predominantly those of non-cirrhotic portal hypertension--either idiopathic portal hypertension with minimal morphological abnormalities, nodular regenerative hyperplasia or in two cases leukaemic infiltration only was noted. Cirrhosis was present in 3/16 cases studied. Both treatment groups developed abnormal liver function tests during the chronic phase, but particularly with progression of the disease. During chronic phase abnormalities were significantly more frequent in those receiving busulphan and thioguanine-alkaline phosphatase (P less than 0.02), transaminases (P less than 0.04), bilirubin (P less than 0.05), multiple abnormalities (P less than 0.01). The development of portal hypertension was often associated with abnormalities of these tests; however, lack of specificity precludes their use as a predictor of subsequent clinical problems. Thioguanine confers no survival advantage in this disease. In view of its hepatotoxicity it should not be used routinely for maintenance of control in chronic phase CML.