Glutathione S-transferases (GSTs) play an important role in the detoxification of xenobiotics in mammals. They catalyze the conjugation of glutathione to a wide range of electrophilic compounds. Phenanthrene 9,10-oxide is a model substrate for GSTs, representing an important group of epoxide substrates. In the present study, combined quantum mechanical/molecular mechanical (QM/MM) simulations of the conjugation of glutathione to phenanthrene 9,10-oxide, catalyzed by the M1-1 isoenzyme from rat, have been carried out to obtain insight into details of the reaction mechanism and the role of solvent present in the highly solvent accessible active site. Reaction-specific AM1 parameters for sulfur have been developed to obtain an accurate modeling of the reaction, and QM/MM solvent interactions in the model have been calibrated. Free energy profiles for the formation of two diastereomeric products were obtained from molecular dynamics simulations of the enzyme, using umbrella sampling and weighted histogram analysis techniques. The barriers (20 kcal/mol) are in good agreement with the overall experimental rate constant and with the formation of equal amounts of the two diastereomeric products, as experimentally observed. Along the reaction pathway, desolvation of the thiolate sulfur of glutathione is observed, in agreement with solvent isotope experiments, as well as increased solvation of the epoxide oxygen of phenanthrene 9,10-oxide, illustrating an important stabilizing role for active site solvent molecules. Important active site interactions have been identified and analyzed. The catalytic effect of Tyr115 through a direct hydrogen bond with the epoxide oxygen of the substrate, which was proposed on the basis of the crystal structure of the (9S,10S) product complex, is supported by the simulations. The indirect interaction through a mediating water molecule, observed in the crystal structure of the (9R,10R) product complex, cannot be confirmed to play a role in the conjugation step. A selection of mutations is modeled. The Asn8Asp mutation, representing one of the differences between the M1-1 and M2-2 isoenzymes, is identified as a possible factor contributing to the difference in the ratio of product formation by these two isoenzymes. The QM/MM reaction pathway simulations provide new and detailed insight into the reaction mechanism of this important class of detoxifying enzymes and illustrate the potential of QM/MM modeling to complement experimental data on enzyme reaction mechanisms.