Abstract: Many outliers and measurement errors exist in the next generation simulation (NGSIM) data. To make the vehicle trajectories more precise and usable for researching, they should be reconstructed before establishing certain models. In this paper, a two-step model was developed:1) The two patterns of outliers were identified by wavelet analysis and physical restricts and modified by 5^th-degree and 3^th-degree Lagrange polynomial interpolation, respectively; 2) The Kalman filter, taking signal energy into account, was conducted to filter the noises in NGSIM data. The performance of the two-step model was supported by the speed curve, acceleration curve, and jerk analysis from the NGSIM database I-80. Finally, this method was implemented to reconstruct all the vehicle trajectories in NGSIM data, and the acceleration distribution indicated that the performance was very good.