The substrate specificity of cytochrome P450, defined as the ability of a compound to promote NAD(P)H and O-2 utilization in the production of either organic or reduced oxygen metabolites, is largely determined by steric and hydrophobic interactions. P450 specificity may therefore be determined by the "fit" of a compound within the active site. A receptor-constrained three-dimensional screening program (DOCK) has been used to select 11 compounds predicted to fit within the P450(cam) active site and 5 compounds predicted to fit within the L244A P450(cam) but not wild-type active site. The 16 compounds were evaluated as P450(cam) substrates by measuring (a) binding to the enzyme, (b) stimulation of NADH and O-2 consumption, (c) enhancement of H2O2 production, and (d) formation of organic metabolites. Seven of the compounds predicted to fit in the active site, and none of the compounds predicted not to fit, were found to be substrates. Compounds predicted to fit very tightly within the active site were poor or non-substrates. The L244A P450(cam) mutant was constructed, expressed, purified, and shown to readily oxidize some of the larger compounds incorrectly predicted to be substrates for the wild-type enzyme. The 5 ligands selected to fit the L244A but not wild-type sites were not detectable substrates, presumably because they fit too tightly into the active site. Retroactive adjustments of the docking program based on an analysis of the docking parameters, particularly variation of the minimum distance allowed between ligand and protein atoms, allow correct predictions for the activity of 15 of the 16 compounds with wild-type P450(cam). The DOCK predictions for the L244A mutant were also improved by changing the minimum contact distances to disfavor the larger compounds. The results indicate that ligands that fill the; active site are marginal or non-substrates. A degree of freedom of motion is required for substrate positioning and catalytic function. if parameters are chosen to allow for this requirement, P450(cam) substrate predictions based on Ligand docking in the active site can be reasonably accurate.