Few tools exist to assess replication of chronic pathogens during infection. This has been a considerable barrier to understanding latent tuberculosis, and efforts to develop new therapies generally assume that the bacteria are very slowly replicating or nonreplicating during latency^,,. To monitor Mycobacterium tuberculosis replication within hosts, we exploit an unstable plasmid that is lost at a steady, quantifiable rate from dividing cells in the absence of antibiotic selection. By applying a mathematical model, we calculate bacterial growth and death rates during infection of mice. We show that during chronic infection, the cumulative bacterial burden—enumerating total live, dead and removed organisms encountered by the mouse lung—is substantially higher than estimates from colony-forming units. Our data show that M. tuberculosis replicates throughout the course of chronic infection of mice and is restrained by the host immune system. This approach may also shed light on the replication dynamics of other chronic pathogens.