The　roadway　section　division　is　the　first　step　to　establish　accident　prediction　model.　Generally　speaking,　there　are　two　methods　in　the　roadway　section　division:　One　is　fixed　length　method,　namely　the　roadway　section　is　divided　into　the　fixed　length,　like　0.5　km　(mile),　1　km　(mile)　and　so　on.　The　other　is　variable　length　method,　in　which　the　section　division　is　usually　conducted　according　to　the　changes　of　the　independent　variables,　e.g.　traffic　volume,　speed　limit　and　geometric　design　elements　like　horizontal　curvature,　vertical　grade,　lane　numbers,　shoulder　width,　and　so　on.　Regarding　the　freeway,　both　two　methods　have　certain　deficiencies.　This　study　proposed　a　new　method　of　freeway　section　division　based　on　the　ordinal　clustering　method,　and　the　clustering　index　is　the　accident　frequency　per　kilometer　distributed　along　the　roadway.　Through　example　analysis,　we　discovered　that　accident　frequency　of　each　segment　obtained　by　this　method　was　obedient　to　certain　probability　distribution,　which　is　beneficial　to　the　model　establishment,　and　this　method　has　a　good　application　value.　Copyright　ASCE　2007.