Aiming　at　the　chiller　process　variables　cannot　be　strictly　obey　the　Gauss　distribution,　and　the　large　number　of　variables　between　the　serious　correlation,　this　paper　describes　a　fault　detection　method　to　detect　the　faults　of　chiller.　Independent　Component　Analysis(ICA)　approach　is　used　to　extract　the　correlation　of　variables　of　chiller　and　reduce　the　dimension　of　measured　data.　A　ICA-based　method　model　is　built　to　determine　the　thresholds　of　statistics　and　calculate　statistics　I-2　and　SPE,　which　are　used　to　check　if　a　fault　occurs　in　chiller.　The　method　is　validated　using　the　laboratory　data　from　ASHRAE　RP-1043　and　compared　with　Principle　Component　Analysis　(PCA).　Results　show　that　the　ICA-based　method　has　better　fault　detection　performance　of　chiller.　It　has　very　good　sensitivity　for　early　fault　and　can　effectively　reduce　the　false　alarm　rate.