Modelling in magnetoencephalography (MEG) and electroencephalography (EEG) is increasingly based on the boundary element method (BEM). We quantify the influence of boundary element discretization on the neuromagnetic and neuroelectric forward and inverse problem for different dipole depths, brain regions and the quasispherical correction. In particular we derive standards for the general use of BEM models in MEG/EEG source localization. For this purpose simulation with single current dipoles, and source reconstructions from somatosensory evoked potentials and magnetic fields were employed. It was found that both local and global discretization influence source reconstruction. Only at a minimum triangle side length of 10 mm was it possible to achieve stable results for MEG and EEG. In order to obtain acceptable errors within the stable region, the ratio of dipole depth to triangle side length must not be less than 0.5. The results obtained from a comparison of the different brain regions indicate that the similarity to spherical geometry might well have an influence on the estimated dipole location, but not so much on its strength. Source reconstruction employing quasispherical correction was found to be the most stable, in particular in the case of coarse BEM discretization.