The application of a new 3-point pharmacophore-fingerprinting package (TOPP, Triplets Of Pharmacophoric Points) to develop QSAR models is discussed. In the CYP2D6 metabolic stability case, these 3D pharmacophoric fingerprints have shown to be as valid as other 3D descriptors and 2D features. Interestingly, it was found in the 3D models that the use of more realistic substrate conformations, by an additional docking step, did not improve the statistical results significantly. A detailed analysis of the generated pharmacophoric hypotheses is consistent with the previously proposed dual interaction mode of substrates within the active site of CYP2D6.