The rational design of selective kinase inhibitors remains a great challenge. Here we describe a physics-based approach to computationally modeling the kinase inhibitor selectivity profile. We retrospectively assessed this protocol by computing the binding profiles of 17 well-known kinase inhibitors against 143 kinases. Next, we predicted the binding profile of the chemotherapy drug mitoxantrone, and chose the predicted top five kinase targets for in vitro kinase assays. Remarkably, mitoxantrone was shown to possess low nanomolar inhibitory activity against PIM1 kinase and to inhibit the PIM1-mediated phosphorylation in cancer cells. We further determined the crystal complex structure of PIM1 bound with mitoxantrone, which reveals the structural and mechanistic basis for a novel mode of PIM1 inhibition. Although mitoxantrone's mechanism of action had been originally thought to act through DNA intercalation and type II topoisomerase inhibition, we hypothesize that PIM1 kinase inhibition might also contribute to mitoxantrone's therapeutic efficacy and specificity.