On　the　basis　of　section　division　and　influence　factors＇　analysis　and　the　collected　data　of　freeway　Accident　Prediction　Model　was　established　with　Generalized　Linear　Regression　method.　Through　comparing　the　regression　results　of　Poisson,　Negative　Binomial,　Zero-inflated　Poisson　and　Zero-inflated　Negative　Binomial　4　kinds　of　statistical　models,　the　Negative　Binomial　distribution　model　was　chosen　to　be　the　best　APM　finally.　The　elasticity　analysis　for　determining　the　marginal　effects　of　the　independent　variables　was　introduced,　and　the　results　indicated　that　environmental　variables　and　traffic　flow　variables　have　great　influence　on　freeway　accident.