In a retrospective study, we compared the use of computer-generated spectral electroencephalographic descriptors with a neurologist's interpretation of raw EEG data. Data were collected from patients undergoing carotid endarterectomy and anesthetized with isoflurane in nitrous oxide and oxygen. The EEG was recorded on magnetic tape during the period immediately before and after occlusion. These tapes were then analyzed off-line using a computer to generate averaged changes in 18 spectral descriptors. A strip-chart of raw EEG before and after the carotid occlusion was interpreted by a neurologist, who assigned changes in the EEG following occlusion to one of four visual inspection groups, depending on the severity of change. A descriptive examination of the distribution of changes in spectral descriptors revealed that no single descriptor adequately reflected the neurologist's interpretation of the raw EEG. Using data from 20 patients, the percent change from preocclusion to postocclusion values for total power, spectral edge frequency, spectral variance, and log spectral variance was examined. Only for the visual inspection group judged by the electroencephalographer to have the most severe ischemic change was there a significant difference in total power and log spectral variance. Although computer-processed EEG devices are of value as trending devices for detecting visual patterns associated with inadequate cerebral perfusion, single descriptors in this study did not consistently reflect a neurologist's diagnosis of ischemia. Single descriptors of spectral EEG analysis may not be sufficient to use as alarm variables in the recognition of cerebral ischemia.