We evaluate a development programme with an important maternal health care component in the Chittagong Hill Tracts of Bangladesh. The region and its mostly indigenous people experienced violent conflict in the past and face a constant risk of recurring conflict. Given this fragile setting, our work differs from conventional impact evaluations by incorporating two conflict indicators: the household’s actual experience of violence and fears of future violence. We find that the intervention undertaken by the United Nations Development Programme (UNDP) succeeded in boosting maternal health care utilisation: antenatal care (ANC) check-ups rose by 18 percentage points, while deliveries at health facilities increased by 23 percentage points. However, household experiences of violent conflict and perceptions of insecurity dampen maternal health care utilisation. Impacts on ANC check-ups are concentrated among households without experience of conflict (19 percentage points compared to 4 percentage points among households with such experience). And households without fears of violence see relatively larger impacts on deliveries at a health facility (37 percentage points compared to 11 percentage points). The programme is successful in raising maternal health care utilisation but its effectiveness has been constrained by the violence experienced and perceived by households.