Fatigue　Driving　is　one　of　the　main　factors　causing　serious　traffic　accidents　and　the　study　of　fatigue　driving　can　contribute　greatly　to　avoiding　road　accidents.　Therefore,　many　researchers　are　making　a　great　effort　on　how　to　detect　driver　fatigue　both　at　home　and　abroad.　This　paper　presents　a　new　method　to　identify　driver　fatigue.　The　driving　behavior　signals,　such　as　gas　pedal,　steering　wheel　angle,　brake　pedals　and　so　on,　and　PERCLOS　are　measured　and　facial　video　are　stored　at　the　same　time.　PERCLOS　and　facial　video　are　used　as　reference,　so　the　feature　vectors　of　the　normal　state　and　fatigue　state　data　can　be　extracted.　The　driving　behavior　data　is　analyzed　at　normal　driving　state　and　fatigued　ones.　The　Distance　Classifier　is　used　to　classify　driving　behavior　signal.　Improved　results　were　achieved　by　using　the　weighted　minimum　distance　classifier　in　comparison　to　the　normalized　Euclidian　Distance　classifier.　Results　show　that　data　of　steering　wheel　angle　and　distance　to　center　of　lane　can　be　used　as　data　source　to　identify　the　driving　state.　The　experimental　results　also　verify　that　the　proposed　method　is　valid.　©　2010　ASCE.