Drug Profile Matching (DPM), a novel virtual affinity fingerprinting method capable of predicting the medical effect profiles of small molecules, was introduced by our group recently. The method exploits the information content of interaction patterns generated by flexible docking to a series of rigidly kept nontarget protein active sites. We presented the ability of DPM to classify molecules excellently, and the question arose, what the contribution of 2D and 3D structural features of the small molecules is to the intriguingly high prediction power of DPM. The present study compared the prediction powers for effect profiles of 1163 FDA-approved drug compounds determined by DPM and ChemAxon 2D and 3D similarity fingerprinting approaches. We found that DPM outperformed the 2D and 3D approaches in almost all therapeutic categories for drug classification except for mechanically rigid structural categories where high accuracy was obtained by all three methods. Moreover, we also tested the predictive power of DPM on external data by reducing the parent data set and demonstrated that DPM can overcome the common screening problems of 2D and 3D similarity methods arising from the presence of structurally diverse molecules in certain effect categories.