BACKGROUND: The univariate study of clinical, electrophysiological, or biochemical variables has been shown to predict the outcome in postanoxic coma in about 50% of patients for each type of variable. Previous studies did not, however, consider the prognostic accuracy of a multivariate approach. METHODS: Sixty patients in coma for more than six hours after cardiac arrest were prospectively examined by means of repeated clinical examinations (including Glasgow coma score (GCS)), EEG, and medianus nerve somatosensory evoked potentials (SEPs). In 16 patients, the early concentrations of serum neuron specific enolase and ionised calcium were also measured. RESULTS: Within the first year after cardiac arrest, 20% of patients made a good neurological recovery; 80% remained in a vegetative state or died. Clinical examination correctly predicted outcome in 58% of patients, SEP in 59%, and EEG in 41%. The combination of clinical examination, SEP, and EEG raised the percentage of correct predictions to 82%, without false pessimistic predictions. Concentrations of serum neuron specific enolase and ionised calcium were of no additional prognostic help. Multivariate regression analysis identified the association of GCS < 8 at 48 hours with abnormal or absent early cortical SEPs as highly predictive of a bad outcome (risk = 97%, 95% confidence interval = 86-99%). CONCLUSION: The combination of GCS at 48 hours, SEP, and if these are non-conclusive, EEG, permits a more reliable prediction of outcome in postanoxic coma than clinical examination alone.