To investigate whether the Glasgow Coma Scale (GCS) can be used reliably and accurately by inexperienced observers, ratings made by observers grouped by level of experience were examined for within-group interobserver disagreements and for discrepancies with scores given by an expert. The GCS was used accurately by experienced and highly trained users, but inexperienced users made consistent errors. The errors were such that they would not be detectable by studies that examine only interobserver agreement, and they were substantial, averaging in some cases more than one point on the four-point and five-point scales of the GCS. Also, the error rates were highest at the intermediate levels of consciousness, for which the detection of changes in condition is vital. The findings support the continued use of the GCS by appropriately qualified personnel, but call into question much of the conventional wisdom about its reliability when used by untrained or inexperienced staff. The findings also suggest that interobserver comparisons are insufficient for establishing the viability of the GCS.