This study was carried out to evaluate the predictive power of a mathematical model called TRISS in identifying avoidable trauma deaths when compared with peer review (PR). The subjects studied comprised the 24 deaths that occurred among 81 severely injured patients (Injury Severity Score greater than or equal to 16) admitted to a trauma unit over a 1 year period. Sixteen patients who had a greater than 50% probability of survival (Ps) calculated by TRISS died. Peer review concluded that five of those were potentially avoidable deaths. Eight deaths occurred in patients with a Ps less than 50%. Peer review regarded those eight as inevitable deaths. The data, when analysed, showed TRISS to have a 100% sensitivity, 42% specificity, 31% positive predictive value and a negative predictive value of 100% when compared with peer review. It is concluded that TRISS tends to over-estimate potentially avoidable death, especially in patients with severe head injury. However it is a good audit filter as it reliably excludes inevitable death.