Fault　detection　and　diagnosis　for　sensor　are　necessary,　which　affect　the　performance　of　the　thermal　power　plant　of　wet　flue　gas　desulphurization　system　seriously.　A　fault　diagnosis　method　using　kernel　principal　component　analysis　(KPCA)　is　proposed　to　affectively　capture　the　nonlinear　relationship　of　the　process　variables,　which　computes　principal　component　in　high　dimensional　feature　space　by　means　of　integral　operators　and　nonlinear　kernel　functions.　The　faults　are　detected　by　calculating　the　statistics　of　the　square　prediction　error　(SPE)　and　identified　by　calculating　the　change　diagram　of　contribution　percentage　of　Hostelling.　At　last,　employing　the　actual　data　from　wet　flue　gas　desulphurization　system　of　Huaneng　Fuzhou　power　plant,　it＇s　proved　effectively　to　detect　and　identify　four　kinds　of　faults,　which　is　the　complete　invalidation　fault,　fixed　bias　fault,　drift　bias　fault　and　precision　degradation　fault.　The　result　shows　the　KPCA　method　has　a　good　performance　in　fault　detection　and　diagnosis.　©　2013　Springer-Verlag　Berlin　Heidelberg.