OBJECTIVE--To determine the frequency of an abnormal bleeding time in patients with cirrhosis and to relate this to known factors that affect primary haemostasis and to the severity of liver disease. DESIGN--Prospective clinical and laboratory study in patients admitted for complications or investigations of liver disease. SETTING--Royal Free Hospital hepatobiliary and liver transplantation unit. SUBJECTS--100 Consecutive inpatients aged 17-74 with various forms of cirrhosis, including alcoholic, biliary, autoimmune, viral, and cryptogenic. At least 10 days had elapsed since any episodes of bleeding, resolution of sepsis, or alcohol intake. No patient was taking any drug known to affect primary haemostasis. MAIN OUTCOME MEASURES--Bleeding time as measured with the Simplate double blade template device. A bleeding time longer than 10 minutes was considered abnormal. Other measures were platelet count, prothrombin time, partial thromboplastin time, packed cell volume, and blood urea, serum bilirubin, and serum albumin concentrations, all measured on each subject at the same time by standard laboratory methods. RESULTS--A weak but significant correlation existed between the bleeding time and the platelet count (rs = 0.483; p less than 0.001). There were significantly lower platelet counts, longer prothrombin times, and higher blood urea and serum bilirubin concentrations in the 42 patients with bleeding times of 10 minutes or more compared with the 58 patients with bleeding times less than 10 minutes. Multiple linear regression analysis showed that the bilirubin concentration as well as the platelet count was independently correlated with the bleeding time. The combination of a platelet count greater than 80 x 10(9)/l and a prothrombin time less than 17 seconds (usually taken as safe limits for performing routine liver biopsy) did not predict a normal bleeding time. Ten of 39 patients fulfilling these criteria had a prolonged bleeding time. CONCLUSIONS--Prolonged bleeding time is common in patients with cirrhosis, even in those with prothrombin times and platelet counts within "safe limits" for invasive procedures. The severity of liver disease as assessed by the bilirubin concentration plays an important part in determining the bleeding time in cirrhosis. The bleeding time should be measured when assessing patients for invasive procedures who have a raised bilirubin concentration or poor hepatic function, even if the platelet count and prothrombin time are considered adequate.