Abstract Skeletonization is a syntactic pattern-recognition method that is applied to gridded data to produce an automatic line drawing, with an associated event catalog. Previous implementations of skeletonization have been tailored for seismic data. Here, we modify that technique to render it more suitable for other types of gridded data, with particular emphasis on aeromagnetic maps. A modification from previous schemes is the use of a two-pass approach, to reduce the effects of an otherwise problematic directional bias that discriminates against events oriented parallel to columns of the grid. The method can be used effectively for filtering aeromagnetic data on the basis of strike direction, event linearity, event amplitude, and polarity. It is based on the delineation of peak-trough pairs (cycles), which are traced throughout the grid to form contiguous events. Cycles and events are characterized by attributes that include amplitude, polarity, and pulse width. Events are further characterized by length, average strike direction, and linearity. The event attributes are stored in a catalog, thus enabling one to perform attribute-based analysis and data filtering. We illustrate our algorithm using two regional aeromagnetic examples from different parts of the Canadian Shield. The first, from the Great Slave Lake shear zone, is dominated by linear anomaly trends produced by faults and mafic dikes. The second, from the Manicouagan region of northeastern Quebec, contains abundant subcircular and arcuate anomaly patterns caused by large intrusive complexes and a meteorite impact structure.