Traditionally, the satellite orbit measurement data are reprocessed by least-square polynomial fit method. Because of the non-linear errors in the satellite tracking, the accuracy of data reprocessing is reduced. The current approach, based on the semi-parametric regression model, holds that the observations can be de-noising via wavelet threshold, and the nonlinear errors can be estimated and removed from the observations. The data reprocessing method with semi-parametric regression model based on the wavelet de-nosing approach is proposed to improve the accuracy of data reprocessing. Finally, the simulation of the reprocessing of a united S-band(USB) satellite orbit measurement data showed that the method could separate the white noise and nonlinear errors, and greatly improve the accuracy of data reprocessing.