We compare three different approaches to scale clearance (CL) from human hepatocyte and microsome CL(int) (intrinsic CL) for 52 drug compounds. By using the well-stirred model with protein binding included only 11% and 30% of the compounds were predicted within 2-fold and the average absolute fold errors (AAFE) for the predictions were 5.9 and 4.1 for hepatocytes and microsomes, respectively. When predictions were performed without protein binding, 59% of the compounds were predicted within 2-fold using either hepatocytes or microsomes and the AAFE was 2.2 and 2.3, respectively. For hepatocytes and microsomes there were significant correlations (P = 8.7 x 10(-13), R(2) = 0.72; P = 2.8 x 10(-9), R(2) = 0.61) between predicted CL(int in vivo) (obtained from in vitro CL(int)) and measured CL(int in vivo) (obtained using the well-stirred model). When CL was calculated from the regression, 76% and 70% of the compounds were predicted within 2-fold and the AAFE was 1.6 and 1.8 for hepatocytes and microsomes, respectively. We demonstrate that microsomes and hepatocytes are in many cases comparable when scaling of CL is performed from regression. By using the hepatocyte regression, CL for 82% of the compounds in an independent test set (n = 11) were predicted within 2-fold (AAFE 1.4). We suggest that a regression line that adjusts for systematic under-predictions should be the first-hand choice for scaling of CL.