This study was set up to test the reliability of the Glasgow Outcome Scale (GOS) when information was obtained from different sources. Eighty assessments were carried out on a group of 58 patients at three different time intervals up to 24 months post-injury. Each assessment consisted of three independently obtained GOS scores for each patient; (i) a score by a research psychologist after interview and neuropsychological testing of the patient; (ii) a score, obtained by post, by the patient's general practitioner (GP), and (iii) a score made by a research worker based on questionnaire information obtained from relatives by post. The agreement between the psychologist's score and that based on the relatives' information was high (r = 0.79 p = 0.001) whereas the correlation between the psychologist's score and that of the GP was low (r = 0.49 p = 0.001). The GPs tended to make overoptimistic assessments and this was most notable at 6 months post-injury when only 50% of the GPs' assessments agreed with those of the psychologist. We have shown that reliability of the GOS varies with the method of obtaining data. Ideally patients should be interviewed and tested by staff who have not been involved in the acute care of the patient. Failing this, information should be obtained from relatives of the patient and used by staff, trained in the use of the GOS, to assign a GOS score.