Bangladesh has made encouraging progress in reducing maternal and neonatal mortality over the past two decades. However, deaths are much higher than in many other countries. The death reporting system to address maternal, neonatal deaths and stillbirths is still poor. Moreover, cause identification for each of the community and facility deaths is not functional. The overall objective of this thesis is to develop, implement and evaluate the Maternal and Neonatal Death Review (MNDR) system in Bangladesh. The study has been conducted in two districts of Bangladesh. A mixed method is used in studies I and II, whereas a qualitative method is used in studies III-V, and cost of MNDR is calculated in study VI. In-depth interviews, focus group discussions, group discussions, participant observations and document reviews are used as data collection techniques. Quantitative data are collected from the MNDR database. In study I, community death notification in the MNDR system was found to be achievable and acceptable at district level in the existing government health system. A simple death notification process is used to capture community-level maternal and neonatal deaths and stillbirths. It was useful for local-level planning by health managers. In study II, death-notification findings explored dense pocket areas in the district. The health system took local initiatives based on the findings. This resulted in visible and tangible changes in care-seeking and client satisfaction. Death numbers in 2012 were reduced in comparison with 2010 in the specific area. In study III, verbal autopsies at community level enabled the identification of medical and social causes of death, including community delays. Deceased family members cordially provided information on deaths to field-level government health workers. The health managers used the findings for a remedial action plan, which was implemented as per causal findings. In study IV, social autopsy highlights social errors in the community, and promotes discussion based on a maternal or neonatal death, or stillbirth. This was aneffective means to  deliver some important messages and to sensitize the community. Importantly, the community itself plans and decides on what should be done in future to avert such deaths. In study V, facility death review of maternal and neonatal deaths was found to be possible and useful in upazila and district facilities. It not only identified medical causes of death, but also explored gaps and challenges in facilities that can be resolved. The findings of facility death reviews were helpful to local health mangers and planners in order to develop appropriate action plans and improve quality of care at facility level. Finally, in study VI, the initial piloting costs required for MNDR implementation were estimated, including large capacity development and other developmental costs. However, in the following year, costs were reduced. Unit cost per activity was 3070 BDT in 2010, but, in the following years, 1887 BDT and 2207 BDT, in 2011 and 2012 respectively.