Crash　risk　analysis　has　been　conducted　to　investigate　the　crash　mechanisms　and　analyze　the　contributing　factors　from　the　aspects　of　traffic　operation　and　geometric　design　perspectives.　Using　Yongtaiwen　freeway　rear-end　crashes,　traffic　flow,　and　road　alignment　data,　this　study　focused　on　investigating　the　heterogeneous　impact　of　different　contributing　factors　on　crash　risk.　Aiming　at　investigating　the　heterogeneous　effects　of　geometric　characteristics　on　crash　risk,　a　Latent　Class　Analysis　(LCA)　method　is　proposed　to　classify　the　samples.　Aiming　at　studying　the　heterogeneous　effects　of　traffic　operational　states　on　crash　risk,　a　Latent　Profile　Analysis　(LPA)　method　is　proposed　to　classify　the　samples.　Based　on　the　classified　homogeneous　subgroups　of　crashes,　logit　models　were　used　to　study　the　relationship　between　microscopic　traffic　flow　variables　and　crash　risk.　In　addition,　a　latent　class　logit　(LCL)　model　was　also　developed　to　simultaneously　analyze　the　heterogeneous　effects　of　geometric　design　features　and　traffic　operational　states　on　crash　risk.　The　modeling　results　showed　that　the　LCA　+　logit　model　has　better　classification　results　and　prediction　performance　than　the　LPA　+logit　model　or　the　LCL　model,　demonstrating　that　the　heterogeneous　effect　of　geometric　design　characteristics　of　the　road　segment　had　better　prediction　accuracy　in　crash　risk　analysis.　©　The　Institution　of　Engineering　&　Technology　2023.