Sterilizing drugs kill Mycobacterium tuberculosis that persists during chemotherapy. Predictive models should mimic the conditions causing persistence in the lesions of cavitary disease, and should grade current anti-tuberculosis drugs according to their sterilizing activity determined in clinical trials. Models should start with old, stationary cultures grown micro-aerophilically. In these, persistent bacilli occur in different populations in which there is no appreciable cell division. Population 1. Grows in liquid culture medium but not on solid medium. Killed by rifampicin. Population 2. Grows on solid culture medium. Killed by rifampicin. Population 3. Grows in liquid medium but not on solid medium. Tolerant of rifampicin. Population 4. Bacilli from Cornell model mice, after treatment with pyrazinamide and isoniazid, cannot grow in liquid or on solid culture medium. Some of these populations are incorporated in models which start with 100-day liquid medium cultures. In model 1 (population 2) the new drug is added and colony counted after 7 days incubation. In models 2 and 3, 100 mg/L rifampicin is added to the 100-day culture when the bacilli lose their ability to grow on solid culture medium (population 3). After re-suspension in rifampicin-free liquid medium for 7 days, the bacilli recover growth on solid medium, when a colony count is done. The new drug is added during incubation with rifampicin in model 3 and at the start of recovery in drug-free medium in model 2. Models 1 and 3 grade isoniazid, rifampicin and pyrazinamide according to their sterilizing activity determined by clinical trials.