Prediction of the binding strength of untested ligands is a central issue in structure-based drug design. In order to rapidly screen large compound databases, simple scoring schemes are often used in target-based virtual screening. The resulting scores often correlate poorly with biological affinities. More rigorous scoring methods, such as MM-PB/SA, correlate better with biological data by considering solvation effects and protein flexibility in the calculation of the binding free energy of a ligand. Here we describe the performance of a modified MM-PB/SA method on 222 Wee1 kinase inhibitors (48 pyridopyrimidine and 174 pyrrolocarbazole derivatives). Docking of these inhibitors into the available Wee1 kinase crystal structure yielded a consistent binding mode, and the derived MM-PB/SA models showed a significant correlation between calculated and experimental data (r(2) values between 0.64 and 0.67). Further study of these models on external test sets of Wee1 kinase inhibitors and structurally related decoys showed that a model based on a single kinase-inhibitor conformation can discriminate the active inhibitors from decoys. We also tested whether the linear interaction energy method with continuum electrostatics (LIECE) yields comparable results to MM-PB/SA and whether the LIECE and MM-PB/SA models can be applied for virtual screening of compound libraries.