The authors evaluated the inter-observer agreement between two experienced clinicians examining 19 unconscious children who were not paralysed or ventilated. Inter-observer reliability was assessed by proportion of agreement, disagreement rate and kappa statistics. Corneal reflexes, pupillary responses to light and motor responses were the most reliably elicited. Reduction of the number of categories improved inter-observer agreement. Some of the disagreement may be attributed to changes in the child's condition during the period of assessment. There was more agreement about the five-category 0-IV scale than the summated Adelaide (10-category) and Jacobi (13-category) scales. The ability of these scales to follow changes in the patient's condition and to predict outcome needs to be evaluated in a prospective trial.