We identified new lead candidates which showed potent dual inhibition against phosphodiesterase-1 and 5 by a ligand-based virtual screening optimized for lead evolution. This virtual screening method, consisting of classification and regression tree analysis using 168 2-center pharmacophore descriptors and 12 macroscopic descriptors, demonstrated a high predictive ability for bioactivity of new chemical compounds. The obtained lead candidates were structurally diverse, although only the structure-activity relationship data of hydroxamic acid derivatives were used to configure the prediction model for the virtual screening.