In a population-based case-control study in the metropolitan areas of Porto Alegre and Pelotas in southern Brazil children dying in infancy from diarrhoea were compared to neighbourhood controls in terms of several social and environmental variables. Factors found to be significantly associated with an increased risk of death from diarrhoea included the non-availability of piped water, the absence of a flush toilet, residence in a poorly built house and household overcrowding. When adjustment was made for confounding variables and the mutual confounding effect of the environmental variables on each other, the only association that remained statistically significant was that with the availability of piped water. The association with poor housing was almost significant (p = 0.052). Compared to those with water piped to their house, those without easy access to piped water were found to be 4.8 times more likely to suffer infant death from diarrhoea (95% confidence interval 1.7 to 13.8) and those with water piped to their plot but not to their house had a 1.5 times greater risk (95% confidence interval 0.8 to 3.0).