The pharmacophore concept is commonly employed in virtual screening for hit identification. A pharmacophore is generally defined as the three-dimensional arrangement of the structural and physicochemical features of a compound responsible for its affinity to a pharmacological target. Given a number of active ligands binding to a particular target in the same manner, it can reasonably be assumed that they have some shared features, a common pharmacophore. We present a growing neural gas (GNG)-based approach for the extraction of the relevant features which we called PENG (pharmacophore elucidation by neural gas). Results of retrospective validation indicate an acceptable quality of the generated models. Additionally a prospective virtual screening for leukotriene A4 hydrolase (LTA4H) inhibitors was performed. LTA4H is a bifunctional zinc metalloprotease which displays both epoxide hydrolase and aminopeptidase activity. We could show that the PENG approach is able to predict the binding mode of the ligand by X-ray crystallography. Furthermore, we identified a novel chemotype of LTA4H inhibitors.