A computer model study on magnetic source imaging from magnetocardiographic data is presented using a cellular automaton model of the entire human heart in the so-called forward problem. A homogeneous boundary element (BE) torso is built up from real magnetic resonance imaging (MRI) cross-sections. The heart model, which has a realistic anatomical shape, is positioned inside the BE torso. In the forward problem the spread of excitation is simulated by applying a modified Huygen's propagation principle. The magnetocardiogram (MCG) and electrocardiogram (ECG) can then be computed following the bidomain theory. From the simulated MCG data, pseudo primary current dipole (PPCD) estimation within the electrically active tissue is performed. The reconstruction space is defined as a surface in the middle of the atrial and ventricular myocardium and septum. The observation space consists of two mutually perpendicular planes closely above the torso surface on the frontal and the left lateral side, respectively. An iterative minimum-norm approach is applied in order to reconstruct PPCD distributions. The errors in PPCD estimation arising from noisy data and regularization algorithms are investigated in more detail.