We ask whether it is more effective to target men, women, or both – with the same intervention in the same context – to improve women’s outcomes when behaviour is governed by gendered social norms. We conduct a cluster-randomized controlled trial of an edutainment intervention in Pakistan – aimed at delaying marriage of adolescent girls, whereas community norms favour early marriage. We find that targeting men, either alone or jointly with women, reduces child marriages in households directly targeted by the intervention. Targeting women, however, either alone or jointly with men, leads to sustained reductions in child marriages at the village level. To rationalize this pattern of results, we build on a model of Bayesian persuasion in the household, where women are more hesitant to deviate from social norms. We extend this by allowing for gender-segregated information transmission from targeted spouses to other households in the village.