Background Indonesia’s national health information systems collect data on maternal deaths but the completeness of reporting is questionable, making it difficult to design appropriate interventions. This study examines the completeness of maternal death reporting by the district health office (DHO) system in Banten Province. Method We used a nested-control study design to compare data on maternal deaths in 2016 from the DHO reporting system and the MADE-IN/MADE-FOR method in two districts and one municipality in Banten Province, with the aim of identifying and characterizing missed deaths in the DHO reporting system. The capture-recapture method was used to assess the magnitude of underreporting of maternal deaths by both systems. Results A total of 169 maternal deaths were reported in the MADE-IN/MADE-FOR study for calendar year 2016 in the three study areas. The DHO system reported 105 maternal deaths for the same period, of which 90 cases were found in both data sources. Capture-recapture analyses suggest that the MADE-IN/MADE-FOR approach identified 92% (95% CI: 87%–95%) of all maternal deaths, while the DHO system captured 57% (95% CI: 50%–64%) of all maternal deaths. Deaths of women who resided in urban areas had four times higher odds (OR 4.3, 95% CI: 1.52–12.3) of being missed by the DHO system compared to deaths among women who lived in rural or remote areas after adjusting for other covariates. Conclusion The DHO reporting system missed approximately half of the maternal deaths in the 3 study areas, suggesting that the DHO system is likely to grossly underestimate the maternal mortality ratio. The DHO reporting system needs to be improved to capture and characterize all maternal deaths.