Abstract: A method to identify modal parameters of large-scale bridge using structure ambient vibration responses is proposed. The method is based on bandwidth restricted empirical mode decomposition (BREDM) and stochastic subspace identification (SSI), first several intrinsic mode functions (IMF) which only represent one frequency component are obtained through EMD, then SSI is used to identify modal parameters. For the features of low natural frequencies and closely spaced mode of large-scale bridge, masking signal is introduced to restrict the EMD bandwidth to eliminate mode mixing. The method is used to identify modal parameters of a large bridge in Gan-Long rail, and the results is compared with Peak Picking method、SSI and finite element analysis, which shows: The method proposed can effectively identify large-scale bridge modal parameters, masking signal solved the problem of mode mixing during EMD, and the false modes in stabilization diagram are restrained.