Various community-based interventions have been proposed to improve maternity care, but hardly any studies have reported the effect of these measures on maternal mortality. In this study, the efficacy of a maternity-care programme to reduce maternal mortality has been evaluated in the context of a primary health-care project in rural Bangladesh. Trained midwives were posted in villages, and asked to attend as many home-deliveries as possible, detect and manage obstetric complications at onset, and accompany patients requiring referral for higher-level care to the project central maternity clinic. The effect of the programme was evaluated by comparison of direct obstetric maternal mortality ratios between the programme area and a neighbouring control area without midwives. Random assignment of the intervention was not possible but potentially confounding characteristics, including coverage and use of other health and family planning services, were similar in both areas. Maternal mortality ratios due to obstetric complications were similar in both areas during the 3 years preceding the start of the programme. By contrast, during the following 3 years, the ratio was significantly lower in the programme than in the control area (1·4 vs 3·8 per 1000 live births, p=0·02). The findings suggest that maternal survival can be improved by the posting of midwives at village level, if they are given proper training, means, supervision, and back-up. The inputs for such a programme to succeed and the constraints of its replication on a large scale should not be underestimated.