OBJECTIVE: To develop and test a clinical protocol for determining post-traumatic amnesia by retrospective questioning. To establish its limits and factors which influence reliability. DESIGN: Two independent assessments using the Rivermead post-traumatic amnesia protocol were undertaken by separate observers on various groups of patients at various time intervals. Analysis investigated the correlations between assessments, the percentage difference between assessments, the number of patients changing category, and the differences between these analyses in the different patient subgroups. Assessments were undertaken both in hospital and in the patients' homes. Four different patient groups were studied. These were group A: 12 inpatients with very severe head injury late after injury; Group B: 40 patients interviewed at home six months after injury; group C: 22 patients interviewed within a few weeks of injury at home; group D: 116 patients interviewed initially within a few weeks and then at six months, on both occasions at home. The Rivermead post-traumatic amnesia protocol involved clinical questioning of the patient to establish how long after injury (in hours/days/weeks) the patient regained continuous day to day memory. All periods of coma were included. Severity was categorised with standard criteria. RESULTS: Overall correlation was good (Spearman's r 0.79), but the correlation was lower for patients with post-traumatic amnesia < 24 hours and when there was a long delay between assessments. In all groups 19%-25% of patients changed categories between assessments, but only 2% changed by two categories. CONCLUSIONS: The assessment of post-traumatic amnesia with the Rivermead post-traumatic amnesia protocol is reasonably reliable. The misclassification rate however, is significant enough that some caution should be taken in individual cases. Other evidence does show post-traumatic amnesia to be valid, and it probably remains the best simple prognostic item available. In clinical practice one should avoid placing too much weight on post-traumatic amnesia alone.