In 31 patients with definite or suspected multiple sclerosis (MS) presenting with a cervical cord syndrome, somatosensory evoked potentials (SEPs) were recorded to median and posterior tibial nerve stimulation, using cephalic and noncephalic reference electrodes. Magnetic resonance imaging (MRI) of the brain and cervical spinal cord was performed, the latter in sagittal and axial views. SEPs were abnormal in 67.7% of patients, whereas MRI showed cervical cord lesions in 74.2% and intracranial lesions possibly involving the somatosensory pathways in 64.5% of cases. A significant correlation was found between abnormalities of cervical (N13) and cortical (N20) potentials following median nerve stimulation with Fz reference and MRI abnormalities involving the ipsilateral or posterior half of the cervical cord, but not the contralateral or anterior half. The N13 potential, recorded from the low cervical region to a supraglottal reference, was most frequently abnormal in patients with MRI lesions at C6 or C7, whereas P14, recorded from the scalp to a clavicle reference, was most often affected by lesions at Cl or the cervicomedullary junction. Abnormalities of the cortical P40 to tibial nerve stimulation were less significantly correlated with cervical MRI lesions. The latency of N20 measured from N9 at the clavicle and the absolute latency of P40 were significantly correlated with the length of MRI abnormalities in the ipsilateral cervical cord. No significant correlation was observed between SEP abnormalities and brain MRI lesions, which it was considered might possibly involve the intracranial somatosensory pathways. It was concluded that (1) the morphological lesions seen in MRI of the cervical cord usually give rise to appropriate electrophysiological deficits, but the occasional finding of a widespread MRI lesion with normal SEP suggests that myelin damage is not the only or the major factor responsible for abnormal MRI signal; and (2) 'clinically silent' lesions apparently involving the radiations and other sensory structures of the brain appear not to give rise to detectable SEP abnormalities, using the methods of the present study.