A ligand-based virtual screening strategy (a combination of pharmacophore model generation, shape-based scoring, and structure clustering analysis) was developed to discover novel SGLT2 inhibitors. The best pharmacophore model, generated from eight glycoside inhibitors, was utilized to virtually screen three chemical databases that led to the identification of three non-glycoside SGLT2 inhibitors. This is the first report of the generation of a pharmacophore model from glycosides that has then been used to discover novel non-glycosides hits.