A predictive relationship between calculated and observed binding affinities for the complexation of ligands to the fibroblast growth factors FGF-1 and FGF-2 based on molecular docking calculations is described. The majority of the ligands examined in this study have high conformational flexibility, and to account for this, multiple conformers were generated for each and subsequently used in flexible docking calculations. Two scoring functions, Gscore and Emodel, were used to quantify the protein:ligand interaction of which the Emodel score showed the best correlation with experimental binding energies. Both scoring functions, however, predicted similar locations for the ligand sulfate groups in the binding site. The van der Waals radii of nonpolar atoms of both the protein and ligand, which modify the effective sizes of both the protein binding site and the ligand, were also systematically altered by factors of 1.0, 0.9, and 0.8 in order to optimize the conditions for predictive docking. Least squares analyses of the Emodel scores against experimental binding energies yielded best r(2) values of 0.91 and 0.83 for FGF-1 and FGF-2, respectively, with slightly lower q(2) values. Optimized scale factor combinations in conjunction with the least squares lines of best fit based on the Emodel function were used to define a predictive model that was tested against ligands not included in the original set. Acceptable predictions of binding affinity were obtained for use in the initial screening of potential leadlike molecules for both FGF-1 and FGF-2.