This paper seeks to discover the magnitude and causality structure of infant mortality--considered a "sentinel even" for quality-of-care indexes in health--in two municipalities of Northeastern Brazil. This is a population based study of the "invoked experimentation" type comparing observed infant mortality with that expected, given a properly functioning maternal and infant care program, allowing for the calculation of a "preventable index" (PDI) for these two municipalities. The preliminary step consisted of an active search and epidemiological investigation of deaths in order to eliminate their underreporting as events. Infant mortality rates in the two areas were relatively low--39 and 44 per thousand live births, respectively--but PDI in both was classified of the order of 40%, thus indicating a causality structure compatible with mortality rates of 100 per thousand. These findings suggest an uneven distribution of deaths, proved by an analytical comparison of the low income population with that of other income brackets (with risk ratios of 8 and 17.6 for total infant mortality and infant mortality from infectious diseases, respectively). PDI proved to be a useful index of preventability of infant deaths, and has the advantage of being simple and easy for health system managers concerned with the quality of health programs devoted to mothers and children to use.