Drunk　driving　is　one　of　the　leading　causes　contributing　to　traffic　crashes.　There　are　numerous　issues　that　need　to　be　resolved　with　the　current　method　of　identifying　drunk　driving.　Driving　behavior,　with　the　characteristic　of　real-time,　was　extensively　researched　to　identify　impaired　driving　behaviors.　In　this　paper,　the　drives　with　BACs　above　0.05%　were　defined　as　drunk　driving　state.　A　detailed　comparison　was　made　between　normal　driving　and　drunk　driving.　The　experiment　in　driving　simulator　was　designed　to　collect　the　driving　performance　data　of　the　groups.　According　to　the　characteristics　analysis　for　the　effect　of　alcohol　on　driving　performance,　seven　significant　indicators　were　extracted　and　the　drunk　driving　was　identified　by　the　Fisher　Discriminant　Method.　The　discriminant　function　demonstrated　a　high　accuracy　of　classification.　The　optimal　critical　score　to　differentiate　normal　from　drinking　state　was　found　to　be　0.　The　evaluation　result　verifies　the　accuracy　of　classification　method.