Modal parameters of power transformer winding are closely related to its vibration feature, dynamic optimal design and winding vibration fault diagnosis. Consequently, it is of importance to accurately identify the modal parameter of power transformer winding. In the paper, an improved Restrained Band Empirical Mode Decomposition (BREMD) based on Particle Warm Optimization (PSO) is proposed to identify the modal parameter of power transformer winding, where a masking signal is introduced in the process of mode decomposition of vibration signal and the PSO algorithm is applied to determine the best frequency of the masking signal. Then the problem of mode mixing in the existed EMD method is restrained effectively and the accuracy of parameter identification of power transformer winding is improved consequently. When compared to the widely used frequency domain method of Polyreference least squares complex frequency domain method of PloyMax, it is seen that the proposed method could obtain the first four natural frequencies and its corresponding damping ratio successfully and accurately with strong anti-interference feature. Therefore, it could be applied effectively to identify the modal parameters of power transformer winding with complex structures.