In clinical trials of two rifamycin antibiotics (rifampin and rifapentine) for treating tuberculosis (TB), patients with cavitary lung lesions did not appear to derive benefit from rifapentine. Rifapentine was found not to outperform rifampin, despite a lower minimum inhibitory concentration against Mycobacterium tuberculosis in mouse models of TB. To understand these findings, we have developed a rabbit model of TB that reliably develops lung cavities with features similar to those of patients with pulmonary cavitary TB. After single or multiple doses of rifampin or rifapentine that produced human-equivalent plasma exposures, rabbits were sacrificed at different time points after dosing. We measured site-of-disease drug pharmacokinetics and tissue drug distribution. We used pharmacokinetic-pharmacodynamic (PK/PD) modeling to estimate drug penetration into different types of tubercular lesions. Both drugs penetrated rabbit lung cellular lesions, as well as the fibrotic cavity wall of cavitary lesions (penetration coefficients ≥1 compared to plasma). For the necrotic liquefied material inside cavitary lesions known as caseum (which contains high numbers of bacteria), the penetration coefficient was 1.0 for rifampin but only 0.25 for rifapentine. When estimates of site-of-disease drug PK were substituted into clinical PK/PD models, the relationship between site-of-action exposure and sputum culture conversion was significant (P < 10−7). We propose that poor penetration of rifapentine into lung cavitary lesions explains, in part, why rifapentine doses required to improve treatment outcomes in two phase 2 clinical trials were four times higher in TB patients with large cavities compared to TB patients without cavitary lung disease.