1. Guangxi Transportation Research Institute Co., Ltd., Nanning 530007, China;
2. Guangxi Key Laboratory of Road Structures and Materials, Nanning 530007, China;
3. College of Geoexploration Science and Technology, Jilin University, Changchun 130026, China

In order to improve the signal-to-noise ratio of the magnetotelluric data, the authors put forward a new denoising method for removing the noises in magnetotelluric data based on the theory of complementary ensemble empirical mode decomposition(CEEMD) and self-adaptive median filtering. By CEEMD decomposing magnetotelluric time-series into different intrinsic mode functions (IMFs) and trend items, and according to the frequency of noise,the authors selectively used self-adaptive median filter to denoise each IMF component to extract useful data from IMFs,then reconstructed the data for signal-noise separation. This method was applied to the measured data, and it is indicated that the method can suppress medium and low frequency noises in magnetotelluric data and inhibit mutation, which improve the signal-to-noise ratio effectively.