The data quality of collected sensing data, which determines the practical value of sensing systems, has been studied in several previous efforts; however, we argue that vehicular ad hoc networks (VANETs), which are a particular application of highly dynamic sensing systems, requires specific treatment to guarantee data quality. In this paper, we design a mechanism, i.e., RD4, which is a role-differentiated cooperative deceptive data-detection and filtering mechanism, to detect the false data in VANETs. RD4 is evaluated using an extended traffic simulator. Three scenarios, i.e., freeway, road construction on a highway, and a traffic light on a local street, are deployed in general. Evaluation results show that the proposed mechanism can achieve more than 90.00% recall and precision rate in most cases.