Abstract: The iterative time-variant filter via the discrete Gabor transform has been found extrmely powerful for wideband and nonstationary signal estimation.It not only is computational efficient, but also offers better SNR.In view of the characteristic of iterative time-variant filter and the nonstationary property of seismic data, a method of the iterative time-variant filter via the discrete Gabor transform for seismic data denoising is proposed.And for the limited frequency band and the distribution reguarities law of real seismic data, a simple method to the mask function is proposed.A great deal of computations of numerical models and real data indicated that the method is efficient.