This　paper　presents　a　study　to　reveal　the　mechanism　of　congestion　impact　on　length-based　vehicle　classification　using　dual-loop　data.　Both　ground-truth　vehicle　trajectory　and　simultaneous　loop　event　data　are　used　to　characterize　the　impact　of　congested　traffic　on　vehicle　classification.　Eight　scenarios　are　synthesized　to　define　the　vehicles＇　stopping　locations　over　two　single　loops.　Under　the　synchronized　traffic,　acceleration　or　deceleration　is　considered　in　the　new　developed　vehicle　classification　under　synchronized　traffic　model.　The　error　of　vehicle　classification　is　reduced　from　33.5%　to　6.7%,　compared　to　the　existing　applied　model.　Under　the　stop-and-go　traffic　condition,　a　stop-on-both-loops-only　(SBL)　was　developed　to　simplify　the　complexity　of　congested　traffic　situation,　resulting　in　the　error　from　235%　to　17.1%.　To　identify　traffic　congestion　states　using　the　dual-loop　data,　an　innovative　method　for　identifying　the　traffic　phases　has　been　proposed　based　relationship　of　speed　and　occupancy,　and　then　successfully　tested　with　the　experimental　data.