Bicycle-motor　vehicle　(BMV)　crashes　have　been　identified　as　a　major　type　of　traffic　accident　affecting　transportation　safety.　In　order　to　determine　the　characteristics　of　BMV　crashes　in　cold　regions,　this　study　presents　an　analysis　using　police-reported　data　from　2015　to　2017　on　BMV　crashes　in　Shenyang,　China.　A　two-stage　approach　integrating　latent　class　analysis　(LCA)　and　the　random　parameter　logit　(RP-logit)　model　is　proposed　to　identify　specific　crash　groups　and　explore　their　contributing　factors.　First,　LCA　was　used　to　classify　data　into　several　homogenous　clusters,　and　then　the　RP-logit　model　was　established　to　identify　significant　factors　in　the　whole　data　model　and　the　cluster-based　model　from　LCA.　The　proposed　two-stage　approach　can　maximize　the　heterogeneity　effects　both　among　clusters　and　within　clusters.　Results　show　that　three　significant　factors　in　the　cluster-based　model　are　obscured　by　the　whole　data　model　in　which　male　cyclists　are　associated　with　a　higher　risk　of　fatality,　especially　in　the　winter.　Additionally,　differences　exist　in　the　exploration　of　factors　due　to　the　characteristics　of　clusters;　thus,　countermeasures　for　specific　crash　groups　should　be　implemented.　This　research　can　provide　references　for　regulators　to　develop　targeted　policies　and　reduce　injury　severity　in　BMV　crashes　in　cold　regions.