In a previous study, an equation with in vivo population binding parameters of carbamazepine and carbamazepine-10, 11-epoxide (CBZ-E) to serum proteins for the relation between unbound and bound serum concentrations was defined. A review by Pynnönen indicates that the average bound/unbound plasma fraction ratio is 3.0 for carbamazepine and 1.0 for CBZ-E. In this study, the ability of equations with in vivo population binding parameters of the previous study (method 1) or with the average bound/unbound plasma fraction ratio of 3.0 of Pynnönen (method 2) to predict the bound serum carbamazepine concentration was retrospectively evaluated using 85 serum samples from 46 patients with epilepsy taking carbamazepine polytherapy. In 21 serum samples from 16 patients, the ability of these equations to predict bound serum CBZ-E concentration was also determined with in vivo population binding parameters from the previous study (method A) or with the average bound/unbound plasma fraction ratio of 1.0 of Pynnönen (method B). Mean prediction error, mean absolute prediction error (MAE), and root mean squared error (RMSE) were calculated for each method, and these values served as a measure of prediction bias and precision. Method 1 showed a bias to overpredict bound serum carbamazepine. The MAE and RMSE were significantly smaller with method 2 (MAE = 2.4 mumol/L; RMSE = 3.2 mumol/L) than with method 1 (MAE = 4.1 mumol/L; RMSE = 4.8 mumol/L). Method 2 was superior to method 1 in terms of accuracy and precision. For bound CBZ-E prediction, method B had a bias to underprediction. The MAE and RMSE were smaller with method A (MAE = 0.581 mumol/L; RMSE = 0.796 mumol/L) than with method B (MAE = 0.724 mumol/L; RMSE = 0.905 mumol/L). Method A was superior to method B in terms of accuracy and precision.