Abstract: This paper presents a new method with adaptive inference capability for prediction of vehicle speed，considering the non-linearity and uncertainty of traffic flow.The first step is to cluster the traffic flow data and to build up the framework of the inference system.The second step is to train and optimize the coefficients with Adaptive Neuro-fuzzy Inference System（ANFIS）.The last step is to simulate and test the system with the MATLAB software.The testing result shows that the system is applicable.