Abstract: NDVI (normalized difference vegetation index)is one of the most important vegetation index, which can reflect vegetation growth and coverage, and it is of great significance to perform real-time monitoring. In this paper, we first take advantage of the amplitude of single-to-noise generated from GPS reflected signal to calculate normalized amplitude, then the characteristics of time series, frequency spectrogram and correlation between the normalized amplitude and NDVI extracted from MODIS products are analyzed. At last, NDVI inversion models with linear regression and BP neural network based on GPS reflection signal are presented. The results show that there are significant annual and seasonal characteristics within the normalized amplitude calculated by GPS reflection signal and NDVI, the correlation coefficient of linear regression is about 0.7, RMSE is 0.05-0.09. Moreover, BP inversion model within consideration of soil moisture is superior to the linear inversion model, the correlation coefficient is increased by about 5%, RMSE is 0.03-0.09. It indicates that NDVI change inversion using GPS reflection signals is feasible, which will provide an alternative approach to monitoring NDVI with high temporal resolution and low cost.