The problem of removing directional trends frequently occurs in the processing of magnetic data and also in the subsequent steps of data interpretation. The so-called corrugations are typical directional trends occurring in levelled data, which may be removed in several ways. Classical techniques are based on high-pass filtering of the data and successively filtering these transformed data with directional cosine filters. Other linear features are due to real sources, such as pipelines in shallow surveys or dike swarms in regional surveys. They should, nevertheless, be considered as noise, due to the fact that their effect is strong and tends to hide the field features related to structures of more interest. We deal with both kinds of problem, presenting the results of a study in an archaeological area of southern Italy. Decorrugation of magnetic field anomalies is performed using a method based on the excellent space–frequency localization properties of wavelet bases, allowing a very sharp filtering of the field along a selected direction. We compare this technique with the classical one in a synthetic case and find that the wavelet decorrugation is simpler and produces low distortion maps. Besides the field decorrugation, the wavelet approach was also shown to be useful in the subsequent enhancement of the measured field. In fact, we show that the wavelet analysis offers a unique framework where various filtering problems (directional, isotropic, global or local as well) may be easily solved. As regards the archaeological case, strong noisy effects from elongated sources (pipelines) were successfully removed in a sharp and local way.