Sybil　attacks　in　Vehicular　Ad-Hoc　Networks　(VANETs)　conduct　malicious　behavior　by　falsifying　and　faking　messages　between　vehicles.　It　poses　a　significant　threat　to　the　safety　of　vehicle　movement.　Meanwhile,　because　Sybil　attacks　often　hide　the　real　identity　of　the　attacker　with　the　help　of　a　legitimate　pseudonym,　making　it　very　difficult　to　detect　them.　Most　existing　detection　schemes　use　a　single　data　source,　which　is　not　enough　to　describe　the　specific　characteristics　of　the　attack　behavior　accurately,　while　their　detection　performance　is　also　affected　by　real　scenario　factors　such　as　traffic　flow　and　attacker　density.　Therefore,　we　propose　a　multi　-source　data　fusion　detection　framework　for　Sybil　attacks　based　on　the　study　of　the　behavior　characteristics　of　Sybil　attacks　and　the　impact　of　the　attacks　on　the　traffic　flow　state.　We　get　basic　safety　messages　data,　map　data　and　sensor　data　and　then　obtain　multi-dimensional　data　fusion　features　from　four　aspects:　spatio-temporal　location　relationship,　traffic　flow　state　change,　vehicle　behavior　characteristics　and　sensor　data　verification,　and　finally　use　machine　learning　classification　model　to　complete　the　detection　of　attack　behavior.　Experimental　results　show　that　our　proposed　attack　detection　framework　is　able　to　locate　the　specific　road　section　where　the　attack　occurred　in　a　realistic　and　complex　traffic　scenario　containing　different　road　types　without　using　trusted　vehicles　as　observation　nodes,　and　has　good　generalization　capability.　the　average　detection　accuracy　of　the　MDFD　framework　for　four　types　of　compound　attacks　is　as　high　as　97.69%.