The S transform is a time-frequency localization technique that bridges the gap between the short-time Fourier transform and wavelet transforms. We propose a new method, designed for application to multiple time series that all have similar time dependence in their spectra, that exploits the properties of the S transform to estimate signal frequency characteristics as a function of time. The method gives an average time-frequency distribution that forms the basis for an adaptive filter for noise attenuation. Using prestack (multichannel) seismic data from a crustal seismic reflection profile in Canada, we show that this technique is particularly effective for determining residual statics corrections for data with a low signal-to-noise ratio.