Aiming at the mode mixing problem in the sifting process of empirical mode decomposition (EMD), a new approach for system identification (SI) based on Hilbert-Huang transform (HHT) and ideal band-pass filter was proposed. First, using Fourier transform, a rough estimation of the natural frequency was obtained from the frequency response function. Then, an ideal band-pass filter was designed, with pass band determined through the half-power bandwidth method. The filter provides zero phase shift within the pass band, and at the same time does not change amplitude spectra. The original response was filtered into a series of narrow band signals through designed filters. Further, narrow band signals were decomposed into modal responses using EMD. The Hilbert transform was applied to each modal response to construct the analytical signal. Finally, the least-square fitting was proposed to identify structural modal and physical parameters. Results show that the half power bandwidth method can realize the quantitative design of the passband. Zero phase shift characteristics of the ideal band-pass filter can better meet the requirements of HHT used in the SI. The combination of the two methods can effectively solve the mode mix problem, eliminate the false modal components and greatly improve SI accuracy.