Evoked potential conduction times in brain stem auditory (BCT) and central somatosensory pathways (CCT) were recorded from 23 normal subjects and 101 patients with severe head injury. Abnormalities in the CCT and the BCT findings correlated with the clinical indices of brain damage (coma score, motor response, pupil response, and spontaneous and reflex eye movements) in the head-injured patients and each correlated with outcome at 6 months from the injury. The CCT in the "best" hemisphere produced the strongest correlation with outcome (P less than 0.001). The correlation of the CCT with outcome was stronger in the 47 patients examined 2 to 3 days after the injury (P less than 0.001) compared to the 34 patients examined within 24 hours after the injury (P less than 0.02). No such difference was noted for the BCT. Serial studies within the first 2 weeks of injury did not show a consistent pattern and repetition of the investigation over this period did not provide any additional information. We used an INDEP-SELECT discriminant analysis program to determine whether information from the evoked potential data could improve prediction of outcome based on clinical data alone. With the addition of the CCT, the predictive accuracy (expressed as the correct classification probability) increased only slightly from 77 to 80%, and the difference was not significant. We conclude that central somatosensory and auditory brain stem conduction times provide useful prognostic information in paralyzed or sedated patients, but when neurological examination is feasible the benefits of evoked potential analysis do not justify the effort involved in data collection.