When a patient resuscitated from cardiac arrest remains unconscious the clinician would like to have a reliable early method for predicting the outcome. The objective of our study was to predict cerebral outcome after cardiac arrest by clinical neurological examination. The data were drawn from an international multicentre controlled clinical trial of thiopentone. Twelve hospitals in nine countries took part. 262 comatose cardiac arrest survivors were followed up for one year. These patients were given advanced life support (American Heart Association guidelines) followed by intensive care to a standardised protocol. Glasgow and Glasgow-Pittsburgh coma scores and their constituent signs were recorded at fixed times. Outcome was taken to be the best cerebral performance at any time during follow-up, and for that purpose we used cerebral performance categories (CPC 1-5) of the Glasgow outcome categories. A poor outcome (CPC 3-5) could be predicted immediately after reperfusion (at entry into the study) with an accuracy ranging from 52% to 84% for various signs and scores. On the third day it was possible to identify severely disabled or permanently comatose survivors without false predictions using both coma scores and several of their constituent variables. The best predictor was absence of motor response to pain. This modelling exercise now needs to be repeated on a new series of patients but the results do suggest that, after 3 days, stringent ethical criteria can be met and used in decision-making about termination of care in comatose cardiac arrest survivors.