OBJECTIVE: Breast-feeding counselling has been identified as the intervention with the greatest potential for reducing child deaths, but there is little experience in delivering breast-feeding counselling at scale within routine health systems. The study aim was to compare rates of exclusive breast-feeding associated with a breast-feeding counselling intervention in which community health agents (CHA) received 20 h of training directed at counselling and practical skills with rates pre-intervention when CHA received 4 h of didactic teaching. DESIGN: Cross-sectional surveys of breast-feeding practices were conducted pre- and post-intervention in random samples of 1266 and 1245 infants aged 0-5.9 months, respectively. SETTING: Recife, Brazil, with a population of 2 million. SUBJECTS: CHA (n 1449) of Brazil's Family Health Programme were trained to provide breast-feeding counselling at home visits. RESULTS: Rates of exclusive breast-feeding improved when CHA were trained to provide breast-feeding counselling and were significantly higher by 10-13 percentage points at age 3-5.9 months when compared with pre-intervention rates (P < 0.05). Post-intervention point prevalence of exclusive breast-feeding for infants aged <4 months was 63% and for those aged <6 months was 50%. CONCLUSIONS: Multifunctional CHA were able to deliver breast-feeding counselling at scale within a routine health service and this was associated with a significant increase in rates of exclusive breast-feeding. The study reinforces the need to focus training on counselling and practical skills; a key component was an interactive style that utilized the knowledge and experience of CHA. The findings are relevant to the call by international organizations to scale up breast-feeding counselling.