The　aim　of　this　paper　is　to　build　a　Hidden　Markov　Model　of　normal　driving　state　of　drivers　from　the　perspective　of　driving　behavior　to　reduce　traffic　accidents　caused　by　driver　fatigue.　Driving　behavior　data　which　drivers　at　normal　driving　state　and　the　fatigues　ones　is　analyzed.　The　vehicle　control　strategies　of　drivers　are　selected　as　hidden　states.　Data　of　driving　behavior　are　denoted　as　observable　output　sequences.　Large　amount　of　driving　behavior　data　is　quantified　with　LBG　VQ　algorithm.　Model　parameters　are　estimated　to　use　Baum-Welch　algorithm.　The　model　is　evaluated　in　driving　state　identification　experiments　using　signals　collected　in　a　driving　simulator　with　forward　algorithm　and　backward　algorithm.　Simulation　results　indicate　that　abnormal　driving　state　of　drivers　can　be　identified.