Background: In low- and middle-income countries, mortality levels are commonly derived from retrospective reports on deceased relatives collected in sample surveys and censuses. These data sources are potentially affected by recall errors. Objective: Using high-quality data collected by the Nouna Health and Demographic Surveillance System (HDSS) in Burkina Faso, we evaluate the reliability of mortality estimates based on the 2006 national census. Methods: We extracted from the census database all records referring to the population under surveillance in the HDSS. Life tables were estimated from recent household deaths reported in the census and compared to those obtained from the prospective mortality data. To evaluate age errors and assess their impact on mortality, we linked census and HDSS records at the individual level for the surviving population and the deceased. Indirect estimates of mortality were also calculated based on the reported survival of children and parents. Results: Life expectancies at birth derived from recent household deaths pointed to a lower mortality than monitored in the HDSS, with a difference of 4 years for men and 8 years for women. Underreporting of deaths among the population aged 60 and above accounted for more than half of these differences. Age errors were small for the surviving population and larger for the deceased, but their effects on mortality estimates were modest. Indirect estimates of child mortality were consistent with the HDSS data, but orphanhood-based estimates were implausibly low. Conclusions: Additional elicitation questions should be asked during the census interviews to improve the collection of data on recent household deaths. Contribution: Mortality rates derived from recent household deaths can seriously underestimate mortality. In Burkina Faso the downward bias in the 2006 census was larger among females and was mostly attributable to underreporting of deaths.