With the increase in the use of shoulder arthroscopy in the past decade, there has been an increased awareness of complications. Reports of the occurrence of transient neuropraxia indicate an incidence of 10%-30%. The recording of somatosensory evoked potentials (SEP) for the study and functional monitoring of the sensory pathway is well accepted as a reproducible method of monitoring peripheral nerve and spinal cord function during surgery. SEPs were recorded during shoulder arthroscopy in 20 patients to monitor the musculocutaneous nerve, ulnar nerve, and either the median or radial nerve. In all 20 cases, abnormal SEPs of the musculocutaneous nerve were demonstrated. In 16 cases, this was produced upon initial joint distention, and in 15 cases, by traction; in 11, by longitudinal traction of greater than or equal to 12 lb, and in six by perpendicular traction of greater than or equal to 7 lb. In 10 patients, there were varying combinations of median, ulnar, and radial nerve involvement. There were two cases of clinical neuropraxia in this series. One resolved in 24 h and one in 48 h. The conclusion is that there is a real potential for neurologic damage during shoulder arthroscopy and that the musculocutaneous nerve is the most vulnerable. Factors responsible include joint distention, excessive traction, and extravasation of fluid. The use of SEPs provides a reliable means for monitoring the neurologic status of the extremity during shoulder arthroscopy.