Intensive research has been performed on computational design of kinase inhibitors using molecular dynamics simulations, docking and quantitative structure-activity relationship (QSAR) analyses, all of which have their own limitations. In this paper, we report the application of proteochemometrics, a ligand-target modeling approach, to the recognition of stable and unstable kinase-inhibitor complexes using support vector machines (SVM) classifiers. The algorithm consists of creating topological autocorrelation descriptors for kinases and inhibitors and then development of SVM models to relate the feature vectors to the stability class (stable or unstable) of hypothetical protein-inhibitor complexes. The approach based on the autocorrelation features was compared with fragment-based approach and the former was found to outperform the later. The final classifier could recognize 82% of data to be stable or unstable using jackknife type of validation and test set prediction. Analysis of substructure classification showed a very homogeneous behavior of the model on the whole target-ligand space. The predictor is available online at http://gibk21.bse.kyutech.ac.jp/AUTOkinI/SVMpredictor.html.