In a population-based study, all infant deaths occurring in a one-year period in the metropolitan areas of Porto Alegre and Pelotas, in southern Brazil, were studied. There were 227 infants who presented diarrhoea during the fatal illness, and in 75% of these diarrhoea was considered to be the underlying cause of death. Acute diarrhoea (< 14 days' duration) accounted for 28% of the deaths, persistent diarrhoea for 62% and dysentery for a further 10%. Approximately one-half of the children with persistent diarrhoea were admitted to a hospital in the first two weeks of the episode. Hospital-acquired infections were likely to have contributed to one- to two-thirds of deaths due to dysentery and persistent diarrhoea. A comparison with neighbourhood controls showed that breast milk provided substantial protection against deaths due to either acute or persistent diarrhoea.