Abstract: In order to improve the short-term multi-step prediction effect of traffic data serials,a dynamic predictability analysis of traffic data serials is first proposed by this paper.Then the characteristics associated indexes of traffic data serials predictability is designed,and the predictability analysis method for traffic data serials is established based on BP neural networks.The traffic data for test are obtained from a real section of urban expressway,the performance of proposed method is compared to different fixed prediction steps.The experimental results indicate that the proposed method can predictability analyze the predictability of traffic data serials on-line,and short-term prediction step was maximized under the premise of maintaining the prediction accuracy.