Abstract: Some research results have shown that the non-stationary characteristics of earthquake ground motion influence the structural responses greatly and to take into account the non-stationary characteristics is essential to the structural analysis. Consequently, it is necessary to describe earthquake ground motion with local spectral density, which is time-dependent and can reflect the non-stationary characteristics. Comparisons of estimating precision and speed were firstly carried out in theory among several local spectral density estimation methods, i.e., wavelet transform, multifilter and short time Fourier transform. Some numerical examples were evaluated to validate the theoretical results. It is concluded that wavelet transform has great advantages on local spectral density estimation due to its higher precision and faster speed than that of the other two methods. Based on wavelet transform, it is convenient and efficient to apply local spectral density to many aspects of earthquake engineering, such as earthquake ground motion modeling, structural random response analysis and reliability evaluation under earthquake excitations.