Phenotypic drug susceptibility testing is the current "gold standard" for detecting Mycobacterium tuberculosis susceptibility to antituberculous drugs. Pyrazinamide is one antituberculous drug for which the correlation between in vitro resistance and clinical outcomes remains unclear. Here we performed latent class analysis (LCA) to develop a consensus gold standard definition of pyrazinamide resistance using three paired standard pyrazinamide resistance assays. We then compared this consensus measure to the 2-month culture results for patients with multidrug-resistant tuberculosis (MDR-TB) who were treated for 2 months with first-line antituberculous drugs before their resistance results were known. Among 121 patients with MDR-TB, 60 (49.6%) were resistant to pyrazinamide by the Wayne method (L. G. Wayne, Am Rev Respir Dis 109:147-151, 1974), 71 (58.7%) were resistant by the Bactec MGIT 960 method, and 68 (56.2%) were resistant by pncA sequencing. LCA grouped isolates with positive results by at least two assays into a category which we considered the "consensus gold standard" for pyrazinamide resistance. The sensitivity and specificity for this consensus gold standard were 82.4% and 92.5%, respectively, for the Wayne method; 95.6% and 88.7%, respectively, for the Bactec MGIT 960 method; and 92.6% and 90.6%, respectively, for pncA sequencing. After we adjusted for other factors associated with poor outcomes, including age, sex, alcohol use, and baseline ethambutol resistance, patients whose isolates were resistant by the LCA-derived consensus gold standard were more likely to be culture positive at 2 months with an odds ratio of 1.95 (95% confidence interval, 0.74 to 5.11), but this result was not statistically significant. These findings underscore the need for improved diagnostics for routine use in programmatic settings.