In　order　to　get　an　early　driving　fatigue　warning　early,　the　EEG　data　for　a　driver　was　recorded　by　an　EEG　apparatus　in　a　driving　simulation　environment.　The　power　spectrum　estimation　was　used　to　establish　the　power　distribution　with　frequency　bands.　Delta　and　alpha　activities　were　referred　to　be　possible　to　predict　driver　fatigue　early.　So　the　prediction　system　was　created　by　the　BP　neural　network,　the　prediction　performance　was　tested　separately　in　three　situations　such　as　the　delta　activity　input,　the　alpha　activity　input　and　the　combination　of　both　input.　The　result　shows　the　prediction　impact　is　the　best　when　the　input　vector　is　the　combination　of　both　and　has　the　basis　on　the　development　of　driving　fatigue　warning　system.