A randomized intervention trial to promote breast-feeding was carried out in southern Brazil. A group of 450 mothers and babies was visited at home 5, 10 and 20 days after birth and compared to a non-visited control group of the same size. Ninety-two per cent of the families visited received the three home visits planned. The evaluation of breast-feeding patterns and reasons for weaning took place 6 months after birth for both groups. Ninety-four per cent of the group visited and 92% of the non-visited controls group were traced on the occasion of the assessment. The intervention increased the duration of breast-feeding (median duration of 120 days in the group visited and 105 days in the controls; p = 0.03) and delayed the introduction of milk bottles (median age of introduction of 90 days in the group visited and 60 days in the controls; p = 0.01). Causes of weaning were classified as underlying, intermediate and immediate. The most common underlying cause of weaning was "the baby cried too much", which suggests that mothers should be taught about normal patterns of infant behaviour in the first weeks of life, particularly the need for crying, and the fact that this not necessarily reflects hunger.