'Hot zones' are areas that have >5% prevalence (or incidence) of multidrug-resistant tuberculosis (MDRTB). We present a new mathematical model (the amplifier model) that tracks the emergence and evolution of multiple (pre-MDR, MDR and post-MDR) strains of drug-resistant Mycobacterium tuberculosis. We reconstruct possible evolutionary trajectories that generated hot zones over the past three decades, and identify the key causal factors. Results are consistent with recently reported World Health Organization (WHO) data. Our analyses yield three important insights. First, paradoxically we found that areas with programs that successfully reduced wild-type pansensitive strains often evolved into hot zones. Second, some hot zones emerged even when MDR strains were substantially less fit (and thus less transmissible) than wild-type pansensitive strains. Third, levels of MDR are driven by case-finding rates, cure rates and amplification probabilities. To effectively control MDRTB in the hot zones, it is essential that the WHO specify a goal for minimizing the amplification probability.