Travel　mode　identification　is　one　of　the　key　issues　in　travel　behavior　analysis.　A　number　of　algorithms　have　been　proposed　to　detect　travel　modes　particularly　by　using　global　positioning　system　(GPS)　data,　whereas　most　algorithms　rarely　consider　traffic　conditions.　To　fill　the　gap,　this　paper　distinguishes　two　representative　travel　modes,　i.e.,　bus　and　car,　by　using　the　random-forests　method,　of　which　the　corresponding　feature　variables　are　examined　under　various　traffic　conditions.　Local　congestion　variables　are　defined　to　reduce　uncertainties　between　bus　and　car.　The　results　indicate　that　the　overall　detection　accuracy　of　the　not-in-congestion　trips　is　as　high　as　94.0%,　and　that　of　in-congestion　trips　is　91.1%,　demonstrating　that　distinguishing　traffic　conditions　using　random　forests　can　reliably　improve　travel　modes　detection　accuracy.　It　is　found　that　distinguishing　local　traffic　conditions　can　further　improve　accuracy.　The　paper　contributes　to　travel　behavior　analysis　and　modeling,　and　the　proposed　method　is　ready　for　a　wide　range　of　transportation　practices,　including　traffic　planning　and　management.