Abstract：Seismic noise suppression using inversion in time-space domain avoids the defeet of predictive filtering and preserves the amplitudes of signals;however，it causes residual noises and some signal loss in seismic section. Having analysed not only the frequency characteristics of signal and random noise but also influence of noise upon filter under different signal/noise ratios，authors studied noise elimination using inversion in wavelet transform domain and then advance frequency divisional inversive noise elimination method. With identical processing parameters，this method brings better noise elimination than others do，so that the seismic section contains more useful informations，thus offering reliable seismic data for interpreters. Theoretical and real application effects prove this method practical.
