OBJECTIVE--To analyse the impact of breast feeding on diarrhoeal disease and survival in children above 1 year of age in Guinea-Bissau, west Africa. DESIGN--A community study of an open cohort followed up weekly by interviews over 15 months. Data on feeding practices, anthropometry, and survival were recorded for three years. SETTING--301 randomly selected houses in a semiurban area in the capital, Bissau. SUBJECTS--849 children aged less than 3 years. MAIN OUTCOME MEASURES--Incidence and duration of diarrhoea, weight for age, and death of a child. RESULTS--The incidence of diarrhoea was higher in weaned children than in partially breast fed children, both in 1 year olds (relative risk 1.41; 95% confidence interval 1.23 to 1.62) and in 2 year olds (1.67; 1.29 to 2.15). The mean duration of an episode of diarrhoea was 5.3 days in breast fed children compared with 6.3 days in weaned children (P = 0.001). Independent of the age of weaning, a similar increase was found in an analysis comparing, for each child, the rate and duration of diarrhoea one month before and one month after weaning. Children with low weight for age were breast fed longer than the better nourished children (P = 0.02). Children aged 12-35 months who were not breast fed had a 3.5 times higher mortality (1.4 to 8.3) than breast fed children. CONCLUSIONS--The beneficial effects of breast feeding are not restricted to infancy. Though children who are partially breast fed after infancy may have a lower state of nutrition than the weaned ones, the benefit in terms of lower morbidity may be more important for child survival in places with a high morbidity from diarrhoea and with high mortality.