The use of molecular modeling in conjunction with site-directed mutagenesis has been extensively used to study substrate orientation within cytochrome P450 active sites and to identify potential residues involved in the positioning and catalytic mechanisms of these substrates. However, because docking studies utilize static models to simulate dynamic P450 enzymes, the effectiveness of these studies is strongly dependent on accurate enzyme models. This study employed a cytochrome P450 3A4 (CYP3A4) crystal structure (Protein Data Bank entry 1W0E) to predict the sites of metabolism of the known CYP3A4 substrate raloxifene. In addition, partial charges were incorporated into the P450 heme moiety to investigate the effect of the modified CYP3A4 model on metabolite prediction with the ligand docking program Autodock. Dehydrogenation of raloxifene to an electrophilic diquinone methide intermediate has been linked to the potent inactivation of CYP3A4. Active site residues involved in the positioning and/or catalysis of raloxifene supporting dehydrogenation were identified with the two models, and site-directed mutagenesis studies were conducted to validate the models. The addition of partial charges to the heme moiety improved the accuracy of the docking studies, increasing the number of conformations predicting dehydrogenation and facilitating the identification of substrate-active site residue interactions. On the basis of the improved model, the Phe215 residue was hypothesized to play an important role in orienting raloxifene for dehydrogenation through a combination of electrostatic and steric interactions. Substitution of this residue with glycine or glutamine significantly decreased dehydrogenation rates without concurrent changes in the rates of raloxifene oxygenation. Thus, the improved structural model predicted novel enzyme-substrate interactions that control the selective dehydrogenation of raloxifene to its protein-binding intermediate.