Evidence on strategies to improve infant and young child feeding in India, a country that carries the world's largest burden of undernutrition, is limited. In the context of a programme evaluation in two districts in Uttar Pradesh, we sought to understand the multiple influences on breastfeeding practices and to model potential programme influence on improving breastfeeding. A cross‐sectional survey was conducted among 1,838 recently delivered women, 1,194 husbands, and 1,353 mothers/mothers‐in‐law. We used bivariate and multivariable logistic regression models to examine the association between key determinants (maternal, household, community, and health services) and breastfeeding outcomes [early initiation of breastfeeding (EIBF)], prelacteal feed, and exclusive breastfeeding (EBF). We used population attributable risk analysis to estimate potential improvement in breastfeeding practices. Breastfeeding practices were suboptimal: EIBF (26.3%), EBF (54%), and prelacteal feeding (33%). EIBF was positively associated with maternal knowledge, counselling during pregnancy/delivery, and vaginal delivery at a health facility. Prelacteal feeds were less likely to be given when mothers had higher knowledge, beliefs and self‐efficacy, delivered at health facility, and mothers/mothers‐in‐law had attended school. EBF was positively associated with maternal knowledge, beliefs and self‐efficacy, parity, and socio‐economic status. High maternal stress and domestic violence contributed to lower EBF. Under optimal programme implementation, we estimate EIBF can be improved by 25%, prelacteal feeding can be reduced by 25%, and EBF can be increased by 23%. A multifactorial approach, including maternal‐, health service‐, family‐, and community‐level interventions has the potential to lead to significant improvements in breastfeeding practices in Uttar Pradesh.