The effects of environmental tobacco smoke (ETS) on breast-feeding patterns are poorly understood, while those of parental smoking on breast-feeding initiation vs. duration have not been clearly delineated. We conducted a prospective, population-based birth cohort study to examine the independent effects of maternal, paternal and ETS on breast-feeding initiation and duration. A total of 6747 Hong Kong Chinese infants were recruited and followed up in 1997-8. We obtained detailed household smoking history and recorded breast-feeding patterns in three follow-up interviews over 9 months. We found that both maternal and paternal smoking were associated with not initiating breast feeding (odds ratio [OR] for ever maternal smoking = 2.51, 95% confidence interval [CI] = 1.63, 3.86; OR for ever paternal smoking = 1.22, 95% CI = 1.08, 1.39). Exposure to ETS in utero and post partum were also related to not starting breast feeding (OR(ETS in utero) = 1.10, 95% CI = 0.99,1.24; OR(ETS post partum) = 1.21, 95% CI = 1.08, 1.36). These effects, however, did not persist for breast-feeding duration of < or = 4 months. Cox proportional hazards modelling confirmed the lack of association between any form of smoking and breast-feeding duration. Our findings suggest that smoking of any kind, during or after pregnancy, is a strong risk indicator for not initiating breast feeding. Smoking as a risk indicator for underlying socio-economic, demographic and psychosocial factors is probably responsible for most of the observed adverse effects, although we cannot rule out direct contributions from pathophysiological mechanisms. Public health strategies directed at these underlying factors should be vigorously pursued to reduce the adverse effects of tobacco on breast feeding and infant health in general.