A review of current techniques and results of monitoring spinal cord function by the intraoperative testing of somatosensory evoked potentials is given. The criteria for an ideal monitoring method are defined: (1) potential alterations occur before the lesion is irreversible, (2) monitoring itself does not harm the patient, (3) there are no false-positive or false-negative results, (4) warning criteria are defined by objective and quantifiable parameters. In recording and stimulation, two different approaches are applied: cortical or spinal recording and peripheral or spinal stimulation. Spinal stimulation techniques are considered more invasive, but an averaged potential is obtained quicker and more reliably by spinal methods. Failure rates in establishing useful monitoring procedures vary between 2.85 and 5%. The N2O-analgesic-relaxant-type of anesthesia is recommended. A precise definition of criteria indicating spinal cord damage has been difficult because of the natural variability of intraoperative evoked potentials. Wide ranges of physiologic, anesthesiologic, and technical and surgical factors have been found to influence intraoperative potential monitoring adversely. The so-called warning criteria drawn from evoked potential changes have so far been set arbitrarily: amplitude reductions of 30-50% for several recordings or at least 15 minutes have mostly been used. It has become clear, however, that warning criteria should be different for healthy or impaired spinal cord function and for cortical and spinal recordings. The value of a lesion-specific spinal cord potential for monitoring remains to be clarified. SEPs are sensitive for demonstrating ischemic changes to the spinal cord, but the limited experience with these lesions does not allow firm conclusions regarding the reversibility of clinical and evoked potential changes in spinal cord ischemia in man. The limited experience with multilevel recording, i.e., simultaneously recording at spinal and cortical level, indicates that epidural recordings are less variable and less failure-prone than cortical recording. Simultaneous multilevel recording also gives more information and allows easier recognition of false-positive or false-negative results. Poor preoperative SEP nearly always preclude useful monitoring. The results obtained so far point out areas where further development is necessary in order to increase the efficacy of this method. Major unsolved problems are (1) definition of warning criteria, (2) incidence of false-positive and false-negative findings, and (3) improvement of data acquisition.(ABSTRACT TRUNCATED AT 400 WORDS)