BACKGROUND: Although the benefits of exclusive breastfeeding for child health and survival, particularly in the post-neonatal period, are established, the independent beneficial effect of early breastfeeding initiation remains unclear. We studied the association between timing of breastfeeding initiation and post-enrolment neonatal and post-neonatal mortality up to 6 months of age, as well as the associations between breastfeeding pattern and mortality. METHODS: We examined associations between timing of breastfeeding initiation, post-enrolment neonatal mortality (enrolment 28 days), and post-neonatal mortality up to 6 months of age (29-180 days) in a large cohort from three neonatal vitamin A trials in Ghana, India, and Tanzania. Newborn babies were eligible for these trials if their mother reported that they were likely to stay in the study area for the next 6 months, they could feed orally, were aged less than 3 days, and the primary caregiver gave informed consent. We excluded infants who initiated breastfeeding after 96 h, did not initiate, or had missing initiation status. We pooled the data from both randomised groups of the three trials and then categorised time of breastfeeding initiation as: at ≤1 h, 2-23 h, and 24-96 h. We defined breastfeeding patterns as exclusive, predominant, or partial breastfeeding at 4 days, 1 month, and 3 months of age. We estimated relative risks using log binomial regression and Poisson regression with robust variances. Multivariate models controlled for site and potential confounders. FINDINGS: Of 99 938 enrolled infants, 99 632 babies initiated breastfeeding by 96 h of age and were included in our prospective cohort. 56 981 (57·2%) initiated breastfeeding at ≤1 h, 38 043 (38·2%) at 2-23 h, and 4608 (4·6%) at 24-96 h. Compared with infants initiating breastfeeding within the first hour of life, neonatal mortality between enrolment and 28 days was higher in infants initiating at 2-23 h (adjusted relative risk 1·41 [95% CI 1·24-1·62], p<0·0001), and in those initiating at 24-96 h (1·79 [1·39-2·30], p<0·0001). These associations were similar when deaths in the first 4 days of life were excluded (1·32 [1·10-1·58], p=0·003, for breastfeeding initiation at 2-23 h, and 1·90 [1·38-2·62], p=0·0001, for initiation at 24-96 h). When data were stratified by exclusive breastfeeding status at 4 days of age (p value for interaction=0·690), these associations were also similar in magnitude but with wider confidence intervals for initiation at 2-23 h (1·41 [1·12-1·77], p=0·003) and for initiation at 24-96 h (1·51 [0·63-3·65], p=0·357). Exclusive breastfeeding was also associated with the lower mortality during the first 6 months of life (1-3 months mortality: exclusive vs partial breastfeeding at 1 month 1·83 [1·45-2·32], p<0·0001, and exclusive breastfeeding vs no breastfeeding at 1 month 10·88 [8·27-14·31], p<0·0001). INTERPRETATION: Our findings suggest that early initiation of breastfeeding reduces neonatal and early infant mortality both through increasing rates of exclusive breastfeeding and by additional mechanisms. Both practices should be promoted by public health programmes and should be used in models to estimate lives saved. FUNDING: Bill & Melinda Gates Foundation through a grant to the WHO.