Abstract: During 12 May 2008 Wenchuan MS8.0 earthquake， Sichuan strong motion network obtained 133 sets of 3-componet acceleration records. This paper processed and analyzed some of the records with different station distance from the fault. Using a clustering algorithm based on the ensemble empirical mode decomposition (EEMD)， this paper effectively extracted the time-frequency distribution of the signal energy， its central frequency， Hilbert energy and the time-frequency characteristics corresponding to the maximum amplitudes， and made a comparative study of EEMD method with Fourier transform and wavelet transform (WT) analysis. This study obtained the following results: For non-linear strong motion records， EEMD can be used to suppress the mode mixing effect existing inempirical mode decomposition (EMD) decomposition; in comparison with Fourier transform and wavelet transform the Hilbert-Huang transform(HHT) marginal spectrum amplitude is larger than the low frequency Fourier spectrum amplitude; different from strong influence of the selected parent wave on WT， HHT can directly isolate inherent mode function (IMF) from the strong motion record， representing inherent characteristics of the original data; Hilbert amplitude spectrum exhibits the concentration of most energy in a certain time and frequency range， whilewavelet spectral energy distribute in a wide frequency range. Therefore， the HHT based on EEMD has obvious advantages in terms of its objectivity and high resolution， being able to extract more time-frequency characteristics of seismic acceleration records.