The published evaluation of methods for identifying preventable trauma deaths contains many unstudied confounding factors. To investigate the reliability of methods for identifying such preventable deaths, we compared three consensus systems using separate five-member general review panels assessing 20 non-central nervous system fatalities: panel A, independent judgments; panel B, discussion of all cases preceding individual judgements; and panel C, independent judgments followed by discussion and equivocal case reassignment. The Kappa concordance index was low for all methods (method A, 0.20; methods B and C, 0.40). Of the 11 deaths judged preventable by at least one panel, only death was judged preventable by all three panels. Consensus agreement (four of five assessors) was 20% for panel A, 45% for panel B, and 10% for panel C (difference between panels B and C,p