Kernel　principal　component　analysis　is　a　type　of　nonlinear　principal　component　analysis,　to　decouple　the　nonlinear　correlation　of　variables　by　using　kernel　functions　and　integral　operators,　and　by　computing　the　principal　components　in　the　high　dimensional　feature　space.　A　method　of　fault　diagnosis　for　dynamic　nonlinear　system　by　dynamic　kernel　principal　component　analysis　is　presented　in　this　paper,　and　the　root　of　fault　causes　is　isolated　by　the　reconstructed　variables　with　nonlinear　least　squares　optimization.　The　simulations　in　the　continuous　stirred-tank　reactor　(CSTR)　indicate　that　the　performances　of　process　monitoring　and　fault　diagnosis　by　this　presented　method　are　superior　to　that　by　kernel　principal　component　analysis.