The use of the median value in seismic data processing has gained in popularity during the past 10 years. Examples include automatic editing of noise spikes, CMP stacking techniques and attenuating the downgoing wavefield in VSP data processing. Directional median filtering can be used to attenuate coherent noise in surface seismic reflection data, and can be applied in a spatially and temporally varying manner across a shot record. The direction of coherent noise events can be estimated using local t-x slant stacks weighted by coherence. A two-dimensional window is moved across the shot record, and at each point on the record slant stacks are taken through the central sample of the window. The range of slowness values used for the local slant stacks are chosen on the basis of the apparent velocity range of both signal and noise. The slowness value that produces the maximum stack is assigned to the central sample of the window. In this way, an instantaneous slowness plot of the shot record is produced. Signal conditioning techniques are used to enhance the instantaneous slowness plot. The instantaneous slowness values for the shot record are stored in a look-up table. The median filter is applied in a spatially and temporally varying way across the record on the basis of the values contained in the look-up table. The median values are subtracted from the shot record to produce the filtered record. The process may be described as an adaptive median beam former. The technique has several advantages over conventional f-k filtering. It can be set up so that it is fully automated, and spikes and noise bursts are not smeared over adjacent traces. Also, the filter has a narrow reject band, so that smearing of small-scale features is reduced. The technique has been applied successfully to seismic reflection data from the Roma Shelf region of the Surat Basin.