Prevailing methods of magnetotelluric (MT) data analysis determine the spectra using variations of the Fourier Transform (FT), which is based on the principle of signal stationarity. However, MT data series are non-stationary random signals that do not meet the basic requirements of conventional methods based on the FT. In recent years, the Hilbert–Huang Transform (HHT) has been regarded as a powerful tool for adaptive analysis of non-linear and non-stationary signals. This paper proposes for the first time the adoption of a new method of analysis for MT data, and focuses on two aspects that are facilitated by applying the HHT. The first aspect is the pretreatment of the MT time series data through selecting MT data subsets, and noise suppression; the other concerns the determination of the impedance and apparent resistivity using the HHT instantaneous spectrum. The conclusion reached through discussion of the first aspect is that the proposed methods can greatly improve the quality of MT data. The conclusion drawn from the second aspect is that the HHT instantaneous spectrum method can overcome the problems described above, and obtain stable and reliable estimation of the impedance tensor, and thus naturally minimise the estimation bias brought about by the non-stationary characteristics of MT data. Therefore, the HHT method is effective in analysing MT data and is able to generate meaningful geological information.