Multiple three-dimensional quantitative structure-activity relationship (3D-QSAR) approaches were applied to predicting passive Caco-2 permeability for a series of 28 inhibitors of rhinovirus replication. Catalyst, genetic function approximation (GFA) with MS-WHIM descriptors, CoMFA, and VolSurf were all used for generating 3D-quantitative structure permeability relationships utilizing a training set of 19 molecules. Each of these approaches was then compared using a test set of nine molecules not present in the training set. Statistical parameters for the test set predictions (r(2) and leave-one-out q(2)) were used to compare the models. It was found that the Catalyst pharmacophore model was the most predictive (test set of predicted versus observed permeability, r(2) = 0.94). This model consisted of a hydrogen bond acceptor, hydrogen bond donor, and ring aromatic feature with a training set correlation of r(2) = 0.83. The CoMFA model consisted of three components with an r(2) value of 0.96 and produced good predictions for the test set (r(2) = 0.84). VolSurf resulted in an r(2) value of 0.76 and good predictions for the test set (r(2) = 0.83). Test set predictions with GFA/WHIM descriptors (r(2) = 0.46) were inferior when compared with the Catalyst, CoMFA, and VolSurf model predictions in this evaluation. In summary it would appear that the 3D techniques have considerable value in predicting passive permeability for a congeneric series of molecules, representing a valuable asset for drug discovery.