Abstract: Floating car technique is an effective means for obtaining aggregated road network operating speeds data,but the processed ones can not reflect the real speed due to the original data collecting errors,which can be attempted to solve by data fusion method.Basing upon three levels of data fusion,three levels are proposed,including filtering data errors by association analysis,taking advantages of the fusion of statistical speed with historic database to deal with inadequate sample representativeness of floating cars,as well as fusing data of floating cars with static detectors to address the problem of the vehicle speed processing under complicated road network circumstances．