We compared scalp somatosensory evoked potential (SEP) recordings by non-cephalic and earlobe reference in 14 healthy subjects and in 5 patients with lesions of the upper cervical cord. In healthy subjects, the scalp to earlobe montage tended to cancel all far-field potentials preceding the scalp P14. On the contrary, the P14 far-field was more difficult to identify in scalp to non-cephalic recordings, because in 12/14 cases it followed another far-field (P13), which was very close in latency to the P14. In 4 patients, the scalp to non-cephalic traces showed a single positive wave (P13/P14 complex) in the P14 latency range. If this complex had been labelled as P14, the somatosensory dysfunction would have been localised above the foramen magnum. On the other hand, the scalp to earlobe recording allowed correct localisation of the lesion since it showed the 'real' and delayed P14 in two patients and no far-field response in the remaining two. Therefore, we propose the use of the scalp to earlobe montage as standard in routine examinations.