This　paper　proposes　a　new　self-tuning　Kalman　filter　with　good　tracking　ability　for　unknown　noise　statistics　and　unknown　abrupt　input　change.　The　new　filter　can　easily　compute　unknown　abrupt　input　and　steady-state　gain　matrix　by　building　up　online　identification　of　ARMAX　innovation　model　in　real　time.　The　simulation　results　of　tracking　a　maneuvering　target　shows　the　effectiveness　of　the　new　method　in　this　paper.　©　2011　IEEE.