According to the Leeds scale for identifying severely head injured patients, death can be predicted with certainty within 12 hours of admission if a patient has a score higher than 13. The withdrawal of treatment from such patients has considerable moral and legal implications. Therefore, to test the reliability of the Leeds scale, it was applied to two patient populations with severe head injuries (479 retrospectively, 131 prospectively). In both groups the scale failed to predict mortality with 100% accuracy: in the first group, 16 of 23 (69.6%) of the patients with a score of more than 13 (and therefore predicted to die) died, and 380 (83.3%) of 456 patients with scores of 13 or less survived; the data for the second group are 6/10 (60%) and 98/121 (81%), respectively. The findings suggest that the Leeds prediction model is not infallible and should be applied cautiously when making decisions about the early termination of care in severely head injured patients.