The literature was systematically reviewed to determine whether a prolonged prothrombin time or elevated international normalized ratio predicts bleeding during invasive diagnostic procedures. MEDLINE and CENTRAL were searched through August 2004, with no language restriction, and reference lists were reviewed. For inclusion, articles must have reported on bleeding in more than five patients with abnormal test results undergoing diagnostic procedures. One trial and 24 observational studies were included. In 2 studies of bronchoscopy, the bleeding rates were similar among those with normal and abnormal tests, with wide confidence intervals (CIs) around the risk differences. During central vein cannulation (3 studies), bleeding rates among those with abnormal tests was unlikely to exceed 2.3 percent. The largest of 3 studies of arteriography found equivalent bleeding rates in patients with and without abnormal tests (risk difference, 0%; 95% CI, -3% to 2%). In the 3 studies of liver biopsy with plugging, bleeding rates were 0, 4, and 5 percent with the upper bounds of the CI as high as 17 percent. In the largest study of transjugular biopsy, the bleeding rate was 1.5 percent (95% CI, 0.3%-4%) in patients with abnormal tests. The highest bleeding rate in the 3 studies of percutaneous liver biopsy was 5.3 percent (95% CI, 1%-13%), similar to the rate in patients with normal test results. There is insufficient evidence to conclude that abnormal test results predict bleeding. Randomized controlled trials should be performed to provide stronger evidence for clinical decision making regarding preprocedure transfusion.