In　practice,　real　data　often　contain　some　outliers　and　usually　they　are　not　easy　to　be　separated　from　the　data　set.　As　sample　variance　and　covariance　are　very　sensitive　to　outliers,　a　novel　algorithm　for　kernel　principal　component　analysis　is　proposed　to　improve　its　robustness　with　the　sample　covariance　by　combined　linear　robust　location　M-estimation　with　kernel　function　to　avoid　adverse　effects　of　outliers.　The　simulation　results　show　that　the　proposed　robust　kernel　principal　component　analysis　can　realize　data　reconstruction　with　outliers　or　general　noises　with　excellent　performance,　high　precision　and　strong　robustness.　ICIC　International　©　2010　ISSN　1881-803X.