Drunk　driving　is　a　serious　threat　to　road　traffic　safety.　It　is　of　great　significance　to　identify　drunk　driving　accurately.　The　drunk　driving　experiment　is　conducted　in　a　driving　simulator.　The　driving　behavior　parameters　under　the　drunk　driving　and　normal　driving　are　collected.　The　steering　wheel　angle　is　selected　as　the　feature　based　on　analysis　of　variance　and　analysis　of　mean.　The　average　sequence　of　steering　wheel　angle　is　calculated　using　a　sliding　data　window.　KNN　and　SVM　are　used　to　identify　the　driver＇s　state.　The　optimum　data　window　and　the　highest　recognition　accuracy　of　the　two　algorithms　under　different　road　alignment　are　obtained.　The　two　classification　methods　are　analyzed.　The　results　show　that　the　recognition　performance　of　the　SVM　is　better　than　that　of　the　KNN.　Data　window　has　a　significant　effect　on　the　performances　of　KNN　and　has　no　significant　effect　on　the　performances　of　SVM.　©,　2015,　Science　Press.　All　right　reserved.