In this paper, we propose a novel source model for a magnetoencephalography (MEG) inverse problem that combines a conventional extended parametric approach and an imaging approach.Our aim is to separately identify a focal current source and background activities spread over the brain.The new source model consists of two terms to represent different spatial characteristics: one is a localized patch source represented with a few parameters based on a mapping from a sphere to the cortex surface,and the other is a distributed source expressed using elemental dipoles on grid points on the cortical surface. We call it a heterogeneous source model, because these two models have not been used simultaneously.Effectiveness of the proposed method is shown via numerical simulations.