Abstract Magnetic surveys of sites contaminated with unexploded ordnance (UXO) locate ferromagnetic sources and discriminate between UXO and some non-UXO (clutter) that can be left in place. High-pass, two-dimensional (2-D) intra-inversion filtering (IIF) is introduced as a partial solution to the vexing problem of overlapping fields of UXO, clutter, and geologic sources. IIF can severely distort target fields while ameliorating effects of overlapping fields. The filter applied to the input data is also applied to unit dipole fields that are used in the inversion algorithm to extract the position, strength, and orientation of the best-fit magnetic dipole. A least-squares converging grid-search inversion, with successive passes using smaller dipole search volumes, is more efficient than an exhaustive search. Between passes, a polynomial is fit to objective function values and the next search volume is centered on the node nearest the minimum of the polynomial. Intra-inversion gradient estimation improves some inversions relative to standard procedures of intra-inversion dc bias estimation of background fields. A flag-node option uses field values at only those nodes nearest to magnetometer stations. To accommodate flag-node data, filter input data windows, data gaps, and/or survey edges, a spatially variable edge-adaptive, gapped, gradient-nulling (EAGGN) filter is derived from a progenitor symmetric, zero-dc filter. It shares with its progenitor the property of nulling any constant-gradient field. EAGGN filter coefficients depend on the distribution of flagged nodes in a filter application window centered on each output location. To overcome problems of swath-overlap noise, single-swath EAGGN filters are applied separately to data from individual swaths of magnetometer arrays. Effectiveness of IIF is demonstrated for anomalies in a magnetic survey of the Blind Test Area of the UXO Standardized Test Site at Aberdeen Proving Ground, Maryland where the U.S. Army Corps of Engineers emplaced UXO simulants and clutter.