Extreme wind such as downburst has strong non-stationary characteristics, which brings challenges to the analyses of wind field characteristics and structural response. Empirically, this non-stationary wind can be divided into a time-varying mean wind and a zero-mean fluctuating wind, which are effective methods to deal with non-stationary wind speed signals. In this paper, the advantages and disadvantages of several common time-varying mean extraction methods are summarized. On this basis, an empirical wavelet transform method based on energy valley search is proposed to extract the time-varying mean wind velocity of downburst. The extracted time-varying mean wind and fluctuating wind are used to analyze the non-stationary response of the top of a high-rise building. The results of two groups of downburst analyses show that compared with empirical mode decomposition and discrete wavelet transform, the time-varying mean wind extracted by energy valley searching empirical wavelet transform is more in line with expectations, and the corresponding structural response is on the safe side.