OBJECTIVES: Indicators of health-system outputs, such as Emergency Obstetric Care (EmOC) density, have been proposed for monitoring progress towards reducing maternal mortality, but are currently underused. We seek to promote them by demonstrating their use at subnational level, evaluating whether they differentiate between a high-maternal-mortality country (Zambia) and a low-maternal-mortality country (Sri Lanka) and assessing whether benchmarks are set at the right level. METHODS: We compared national and subnational density of health facilities, EmOC facilities and health professionals against current benchmarks for Zambia and Sri Lanka. For Zambia, we also examined geographical accessibility by linking health facility data to population data. RESULTS: Both countries performed similarly in terms of EmOC facility density, implying this indicator, as currently used, fails to discriminate between high- and low-maternal-mortality settings. In Zambia, the WHO benchmarks for doctors/midwives were met overall, but distribution between provinces was highly unequal. Sri Lanka overshot the suggested benchmarks by three times for midwives and over 30 times for doctors. Geographical access in Zambia--which is much less densely populated than Sri Lanka--was poor, less than half the population lived within 15 km of an EmOC facility. CONCLUSIONS: Current health-system output indicators and benchmarks on EmOC need revision to enhance discriminatory power and should be adapted for different population densities. Subnational disaggregation and assessing geographical access can identify gaps in EmOC provision and should be routinely considered. Increased use of an improved set of output indicators is crucial for guiding international efforts towards reducing maternal mortality.