To determine the value of EEG in predicting prognosis, the recordings of 30 neonates were re-examined using a specially devised method of visual rating which permitted quantitative assessment of 13 EEG items. Seven clinical variables were also coded. The data were submitted to computer discriminant function analysis, to determine whether the outcome could be predicted and which EEG variables contributed most effectively. Marked asymmetries, many slow transients, and a wide range of variability of suppression durations were the main indicators. The state of consciousness during the recording and the gestational age were the clinical variables which were useful. It was possible to predict outcome using this technique in 28 of the 30 patients. The same procedure was applied to serial recordings with equal value. The combination of visual ratings carried out, without reference to clinical information and requiring no subjective statement as to normality or abnormality of the features in the recording, seems to offer a contribution in the difficult field of neonatal EEG.