We investigated the influence of induced fit of the androgen receptor binding pocket on free energies of ligand binding. On the basis of a novel alignment procedure using flexible docking, molecular dynamics simulations, and linear-interaction energy analysis, we simulated the binding of 119 molecules representing six compound classes. The superposition of the ligand molecules emerging from the combined protocol served as input for Raptor, a receptor-modeling tool based on multidimensional QSAR allowing for ligand-dependent induced fit. Throughout our study, protein flexibility was explicitly accounted for. The model converged at a cross-validated r(2) = 0.858 (88 training compounds) and yielded a predictive r(2) = 0.792 (26 test compounds), thereby predicting the binding affinity of all compounds close to their experimental value. We then challenged the model by testing five molecules not belonging to compound classes used to train the model: the IC(50) values were predicted within a factor of 4.5 compared to the experimental data. The demonstrated predictivity of the model suggests that our approach may well be beneficial for both drug discovery and the screening of environmental chemicals for endocrine-disrupting effects, a problem that has recently become a cause for concern among scientists, environmental advocates, and politicians alike.