Traditional　MKPLS　method　conducts　the　covariance　matrix　decomposition　of　the　data　matrix,　and　some　useful　high-order　statistics　are　not　used,　which　will　cause　the　loss　of　the　useful　data　information　in　the　feature　extraction　process　and　lead　to　poor　fault　recognition　performance.　Aiming　at　this　issue,　a　multi-way　statistics　pattern　analysis　kernel　partial　least　squares　method　(MSPAKPLS)　is　proposed,　which　combines　the　statistics　pattern　analysis　(SPA)　with　multi-way　kernel　partial　least　squares　(MKPLS).　This　method　first　introduces　a　slide　window　technique　to　construct　different　order　statistics　of　the　data　sample;　the　data　are　mapped　from　the　original　data　space　into　the　statistic　sample　space,　then　the　kernel　function　is　used　to　map　the　data　from　the　statistic　sample　space　into　the　high　dimensional　kernel　space,　and　the　PLS　analysis　and　product　quality　prediction　are　conducted.　At　last,　this　method　was　applied　in　the　industrial　penicillin　fermentation　process　and　compared　with　some　conventional　methods;　the　results　show　that　the　proposed　method　has　better　monitoring　and　prediction　performance.