Regional compilations of airborne magnetic data are becoming more common as national databases grow. Grids of the magnetic survey data are joined together to form geological province‐scale or even continental‐scale compilations. The advantage of these compilations is that large tectonic features and geological provinces can be better mapped and interpreted. We take a holistic approach to the joining of survey grids. The leveling of the grids into a regional compilation is treated as a single inverse problem. We use the weighted least‐squares method to find the best adjustment for each survey grid such that the data value differences in the grid overlap areas are minimized. The method spreads any inconsistencies between grids among all of the grid overlap areas and minimizes the introduction of long‐wavelength errors into the composite grid. This is an improvement on the conventional approach of joining grids sequentially. A comparison of leveled data over Western Australia with diurnally‐corrected long aeromagnetic traverses shows long‐wavelength errors of about 200 nT over distances of more than 5000 km. This is an improvement on the sequential grid‐joining method, which gives errors of about 450 nT over the same distance. The application of the method to a smaller area covered by good quality surveys resulted in long‐wavelength errors of about 30 nT over a distance of 1200 km. This is within the estimated accuracy of the original survey measurements. The new method is also fast—what used to take many weeks of effort can now be achieved in a matter of hours.