Glutathione S-transferases (GSTs) are an important class of phase II (de)toxifying enzymes, catalyzing the conjugation of glutathione (GSH) to electrophilic species. Recently, a number of cytosolic GSTs was crystallized. In the present study, molecular modeling techniques have been used to derive a three-dimensional homology model for rat GST 4-4 based upon the crystal structure of rat GST 3-3, both members of the mu class. GST 3-3 and GST 4-4 isoenzymes share a sequence homology of 88%. GST 4-4 distinguishes itself from GST 3-3 in being much more efficient and stereoselective in the nucleophilic addition of GSH to epoxides and alpha,beta-unsaturated ketones. GST 3-3, however, is much more efficient in catalyzing nucleophilic aromatic substitution reactions. In this study, several known substrates of GST 4-4 were selected and their GSH conjugates docked into the active site of GST 4-4. GSH conjugates of phenanthrene 9(S),10(R)-oxide and 4,5-diazaphenanthrene 9(S),10(R)-oxide were docked into the active site of both GST 3-3 and GST 4-4. From these homology modeling and docking data, the difference in stereoselectivity between GST 3-3 and GST 4-4 for the R- and S-configured carbons of the oxirane moiety could be rationalized. The data acquired from a recently derived small molecule model for GST 4-4 substrates were compared with the results of the present protein homology model of GST 4-4. The energy optimized positions of the conjugates in the protein model agreed very well with the original relative positions of the substrates within the substrate model, confirming the usefulness of small molecule models in the absence of structural protein data. The protein homology model, together with the substrate model, will be useful to further rationalize the substrate selectivity of GST 4-4, and to identify new potential GST 4-4 substrates.