Background: Madagascar today has one of the highest life expectancies in sub-Saharan Africa, despite being among the poorest countries in the continent. There are relatively few detailed accounts of the epidemiological transition in this country due to the lack of a comprehensive death registration system at the national level. However, in Madagascar's capital city, death registration was established around the start of the 20th century and is now considered virtually complete. Objective: We provide an overview of trends in all-cause and cause-specific mortality in Antananarivo to document the timing and pace of the mortality decline and the changes in the cause-of-death structure. Design: Death registers covering the period 1976–2012 were digitized and the population at risk of dying was estimated from available censuses and surveys. Trends for the period 1900–1976 were partly reconstructed from published sources. Results: The crude death rate stagnated around 30‰ until the 1940s in Antananarivo. Mortality declined rapidly after the World War II and then resurged again in the 1980s as a result of the re-emergence of malaria and the collapse of Madagascar's economy. Over the past 30 years, impressive gains in life expectancy have been registered thanks to the unabated decline in child mortality, despite political instability, a lasting economic crisis and the persistence of high rates of chronic malnutrition. Progress in adult survival has been more modest because reductions in infectious diseases and diseases of the respiratory system have been partly offset by increases in cardiovascular diseases, neoplasms, and other diseases, particularly at age 50 years and over. Conclusions: The transition in Antananarivo has been protracted and largely dependent on anti-microbial and anti-parasitic medicine. The capital city now faces a double burden of communicable and non-communicable diseases. The ongoing registration of deaths in the capital generates a unique database to evaluate the performance of the health system and measure intervention impacts.