The hypothesis that focal scalp EEG and MEG interictal epileptiform activity can be modelled by single dipoles or by a limited number of dipoles was examined. The time course and spatial distribution of interictal activity recorded simultaneously by surface electrodes and by electrodes next to mesial temporal structures in 12 patients being assessed for epilepsy surgery have been studied to estimate the degree of confinement of neural activity present during interictal paroxysms, and the degree to which volume conduction and neural propagation take part in the diffusion of interictal activity. Also, intrapatient topographical correlations of ictal onset zone and deep interictal activity have been studied. Correlations between the amplitudes of deep and surface recordings, together with previous reports on the amplitude of scalp signals produced by artificially implanted dipoles suggest that the ratio of deep to surface activity recorded during interictal epileptiform activity on the scalp is around 1:2000. This implies that most such activity recorded on the scalp does not arise from volume conduction from deep structures but is generated in the underlying neocortex. Also, time delays of up to 220 ms recorded between interictal paroxysms at different recording sites show that interictal epileptiform activity can propagate neuronally within several milliseconds to relatively remote cortex. Large areas of archicortex and neocortex can then be simultaneously or sequentially active via three possible mechanisms: (1) by fast association fibres directly, (2) by fast association fibres that trigger local phenomena which in turn give rise to sharp/slow waves or spikes, and (3) propagation along the neocortex. The low ratio of deep-to-surface signal on the scalp and the simultaneous activation of large neocortical areas can yield spurious equivalent dipoles localised in deeper structures. Frequent interictal spike activities can also take place independently in areas other than the ictal onset zone and their interictal propagation to the surface is independent of their capacity to trigger seizures. It is concluded that: (1) the deep-to-surface ratios of electromagnetic fields from deep sources are extremely low on the scalp; (2) single dipoles or a limited number of dipoles are not adequate for surgical assessment; (3) the correct localisation of the onset of interictal activity does not necessarily imply the onset of seizures in the region or in the same hemisphere. It is suggested that, until volume conduction and neurophysiological propagation can be distinguished, semiempirical correlations between symptomatology, surgical outcome, and detailed presurgical modeling of the neocortical projection patterns by combined MEG, EEG, and MRI could be more fruitful than source localization with unrealistic source models.