Several derivatives out of a series of antifungal agents exhibited a good inhibitory potency against aromatase as well as a fairly good selectivity toward CYP17, even if lacking H-bond accepting substituents. Their common structural feature is a flexible backbone that did not fit into previously reported CYP19 models. Thus, a ligand-based approach was exploited to develop a novel statistically robust, self-consistent and predictive 3D-QSAR model herein proposed as a helpful tool to design new aromatase inhibitors. © 2008 Elsevier Ltd. All rights reserved.