The association between infant feeding habits and infant mortality from diarrhea was investigated in a population-based case-control study in two urban areas in southern Brazil during 1985. Each of 170 infants who died due to diarrhea was compared with two neighborhood controls. After allowance was made for confounding variables, infants who received powdered milk or cow's milk, in addition to breast milk, were at 4.2 times (95% confidence interval (CI) 1.7-10.1) the risk of death from diarrhea compared with infants who did not receive artificial milk, while the risk for infants who did not receive any breast milk was 14.2 times higher (95% CI 5.9-34.1). Similar results were obtained when infants who died from diarrhea were compared with infants who died from diseases that were presumed to be due to noninfectious causes. Each additional daily breast feed reduced the risk of diarrhea death by 20% (95% CI 2-34%), but the increase in risk associated with each bottle feed was not significant after allowance was made for the number of breast feeds. The only other consumption variable associated with diarrhea mortality was the frequency with which tea, water, or juice were drunk with each feed (increase in risk, 42% (95% CI 4-93%]. The odds ratios associated with nonbreast milk were highest in the first two months of life. Possible biases were investigated, including the interruption of breast-feeding as an early consequence of the terminal illness, but the strong protective effect of breast-feeding persisted after these adjustments.