The field continuation procedure by Ivan (1994) providing potential field continuation between arbitrary surfaces has been implemented for use of unevenly scattered rather than gridded data. Errors induced by under-sampling are investigated by synthetic case studies. The basic assumption of dipole magnitudes linearly varying on the triangle facets of the polyhedral topography approximation turns out to limit the accuracy of Ivan’s algorithm. The method is applied to combine aeromagnetic data acquired by different partly overlapping surveys in order to remove unknown biases and to refer all data to a common surface with constant topography clearance.