Mycobacterium tuberculosis, which causes tuberculosis, remains a major human public health threat. This is largely due to a sizeable reservoir of latently infected individuals, who may relapse into active disease decades after first acquiring the infection. Furthermore, patients have a very slow response to treatment of active disease. Latency and antibiotic tolerance are commonly taken as a proxy for dormancy, a stable nonreplicative state. However, latency is a clinical term that is solely defined by a lack of disease indicators. The actual state of the bacterium in human latency is not well understood. Here we evaluate the results of several in vitro models of dormancy and consider the applicability of various animal models for studying aspects of human latency and resistance to killing by antibiotics. Furthermore, we propose a model for the initiation of dormancy and resuscitation during infection.