Transportation　is　an　important　factor　that　affects　energy　consumption,　and　driving　behavior　is　one　of　the　main　factors　affecting　vehicle　fuel　consumption.　The　purpose　of　this　paper　is　to　improve　fuel　consumption　monitoring　databases　based　on　mobile　phone　data.　Based　on　the　mobile　phone　terminals　and　on-board　diagnostic　system　(OBD)　installed　in　taxis,　driving　behavior　data　and　fuel　consumption　data　are　extracted,　respectively.　By　matching　the　driving　behavior　data　collected　by　a　mobile　phone　with　the　fuel　consumption　data　collected　by　OBD,　the　correlation　between　driving　behavior　and　fuel　consumption　is　explored,　so　that　vehicle　fuel　consumption　could　be　predicted　based　on　mobile　phone　data.　The　fuel　consumption　prediction　models　are　built　using　back　propagation　(BP)　neural　network,　support　vector　regression　(SVR),　and　random　forests.　The　results　show　that　the　average　speed,　average　speed　except　for　idle　(ASEI),　average　acceleration,　average　deceleration,　acceleration　time　percentage,　deceleration　time　percentage,　and　cruising　time　percentage　are　important　indicators　for　fuel　consumption　evaluation.　All　three　models　could　predict　fuel　consumption　accurately,　with　an　absolute　relative　error　less　than　10%.　The　random　forest　model　is　proved　to　have　the　highest　accuracy　and　runs　faster,　making　it　suitable　for　wide　application.　This　method　lays　a　foundation　for　monitoring　database　improvement　and　fine　management　of　urban　transportation　fuel　consumption.