SummarySource generated surface waves are a long-standing issue in land seismic data processing. Indeed they represent one of the main obstacles to the development of time-lapse surveys on land. The present work compares the results of coherent noise filtering obtained on real data sets by using the popular 3D wavenumber filter with the results obtained by using 3D filters designed with wavelet transform techniques. Practical aspects of sampling irregularities are discussed, with respect to the impact on signal preservation. The (f-k x -k y ) filter is a robust, well understood technique, which provides, in general, good results in the cross spread domain with computational efficiency. Wavelet transform tools can deliver superior results in some instances, but they are currently less easy to handle in actual projects. They do, however, provide non linear filtering capabilities, and, in some instances, the option of applying the filter in the shot point domain, avoiding computationally expensive cross spread sorting