Reliable estimation of the local spatial extent of neural activity is a key to the quantitative analysis of MEG sources across subjects and conditions. In association with an understanding of the temporal dynamics among multiple areas, this would represent a major advance in electrophysiological source imaging. Parametric current dipole approaches to MEG (and EEG) source localization can rapidly generate a physical model of neural current generators using a limited number of parameters. However, physiological interpretation of these models is often difficult, especially in terms of the spatial extent of the true cortical activity. In new approaches using multipolar source models [3, 5], similar problems remain in the analysis of the higher-order source moments as parameters of cortical extent. Imagebased approaches to the inverse problem provide a direct estimate of cortical current generators, but computationally expensive nonlinear methods are required to produce focal sources [1,4]. Recent efforts describe how a cortical patch can be grown until a best fit to the data is reached in the leastsquares sense [6], but computational considerations necessitate that the growth be seeded in predefined regions of interest. In a previous study [2], a source obtained using a parametric model was remapped onto the cortex by growing a patch of cortical dipoles in the vicinity of the parametric source until the forward MEG or EEG fields of the parametric and cortical sources matched. The source models were dipoles and firstorder multipoles. We propose to combine the parametric and imaging methods for MEG source characterization to take advantage of (i) the parsimonious and computationally efficient nature of parametric source localization methods and (ii) the anatomical and physiological consistency of imaging techniques that use relevant a priori information. By performing the cortical remapping/imaging step by matching the multipole expansions of the original parametric source and the equivalent cortical patch, rather than their forward fields, we achieve significant reductions in computational complexity.