Phase 2 clinical trials for tuberculosis (TB) treatment require reliable culture methods to determine presence or absence of Mycobacterium tuberculosis (Mtb) over the course of therapy, as these trials are based primarily on bacteriological endpoints. We evaluate which of 5 solid media is most reliable: Lowenstein–Jensen (LJ) egg-base medium and 4 Middlebrook agar media (nonselective 7H10 and 7H11 and selective 7H10 and 7H11). We analyze 393 specimens from 50 HIV-negative Ugandan adults with newly-diagnosed, pulmonary TB and high acid-fast bacillus smear grade. Specimens were collected every 2–4 weeks during the first 12 weeks of therapy. We compare the results for each culture to 2 composite reference standards—one that was deemed positive if any solid culture was positive for Mtb and another based on latent-class analysis. Both reference standards established that the 2 selective Middlebrook media most reliably determine the presence or absence of Mtb (P < 0.003), largely because of their lower contamination rates. We also showed that results on Middlebrook media were similar to each other, while LJ was most frequently discordant. Contaminated results appeared more likely to be truly negative than to harbor undetected Mtb.