This paper analyzes the validity and reliability of a method proposed by HERMAN et al [8] used to classify avoidable neonatal deaths. This method is based on a list of amenable medical conditions with an a priori decision about the avoidance of deaths. The results obtained using this method are compared to those derived from the discussion of individual cases by a committee created ex profeso. The population under study includes all neonatal deaths occurred at a third level hospital in Mexico City, from January 1, 1987 to July 31, 1994 (n = 1337). Only 56% of neonatal deaths could follow HERMAN's classification (n = 749). Poor concordance (Cohen's Kappa = 0.30) between the two methods was found. A high proportion of deaths (72.7%) was classified ambiguously (as possibly preventable), and also a considerable proportion of deaths could not be classified (44%). High sensitivity (96%) was found for the small percentage of cases in which avoidance was determined by the method under assessment (15%). A priori classification is useful for developing rough quality indicators at the regional level but not at hospital settings.