Résumé : In brain topographic mapping, the putative location and orientation in the head space of neural generators are currently inferred from the features of negative and positive scalp potential fields. This procedure requires the use of a fairly neutral reference. The frequently advocated average reference creates problems because its effect is not merely to change a (steady) zero reference level, but to dynamically zero-center all scalp potentials at each latency. Ghost potential fields are thus created at the latencies for which the integral of scalp recorded potentials differs from zero. These distortions of brain mapping have been analyzed with a true 3-shell head model in conjunction with the emulation of SEP components. In the head model, surface potential fields generated by dipoles or dipole sheets of various depths and orientations were computed either over the north hemisphere, so as to emulate scalp recorded SEP components, or over the entire equivalent head sphere. The spurious effects of the average reference are shown to occur because it is computed from a limited number of (scalp) electrodes which fail to survey the bottom half of the head.