Abstract: The long-period noise in magnetotelluric signals mainly exists in the low-frequency and very low-frequency segments of the signal. This type of noise has large amplitude and strong randomness,which can completely submerge the effective electromagnetic signal and cause serious deviations in the low-frequency inversion results. In order to effectively remove the long-period noise in the magnetotelluric signal,this paper combines the characteristics of Variational Mode Decomposition (VMD) and multi-resolution analysis,and proposes a multi-resolution VMD algorithm:first,perform multi-layer VMD processing on the low-frequency Intrinsic Mode Function (IMF) of the signal to improve the time-frequency resolution of the low-frequency and very low-frequency segments; then according to the time-frequency characteristics of the noise,the IMF components containing long-period noise are accurately selected; and finally removed these IMF components achieve the purpose of removing long-period noise. Using this algorithm to process the simulated and measured magnetotelluric signals separately,the results show that:the multi-resolution VMD algorithm can significantly improve the time-frequency resolution of long-period noise,and this algorithm can effectively remove the long-period side of the magnetotelluric signal. Wave and triangle wave noise,the periodicity and smoothness of the time-domain waveform after denoising are significantly improved,while the low-frequency effective components of the signal are preserved,and the apparent resistivity and phase diagram curves of the low-frequency segment have been significantly optimized.