Pyrazinamide, a first-line drug with remarkable sterilizing activity, plays an important role in the treatment of tuberculosis, especially in multi-drug resistant strains. Pyrazinimide use, however, is complicated by its side-effects and that there is no reliable drug susceptibility test. Resistance to pyrazinamide is largely driven by variants in pyrazinamidase (pncA), responsible for drug activation, but large genetic diversity and heterogeneity has hindered the development of a molecular diagnostic test. Our objective was to use information from the proteins 3D structure to accurately identify resistance variants in pncA. To achieve this, we curated 617 pncA non-synonymous single nucleotide variants (nsSNV’s) with experimental and clinical information on pyrazinamide susceptibility. The molecular consequences of these variants were assessed using the mCSM platform, which provided insights into changes in protein stability, conformation, and interactions. Using these structural and biophysical effects, we were able to correctly classify variants as either susceptible or resistant with an accuracy of 77%. Our model was validated against a non-redundant blind test of clinically documented resistance mutations, achieving 100% accuracy, and the mutations recently reported in the CRyPTIC dataset. Applying this structural analysis to a novel clinical variant from a Victorian tuberculosis patient showed pyrazinamide treatment would not be effective and led to its discontinuation, the first use of structural information to guide clinical resistance detection. Therefore, while current WHO recommendations suggest continuing treatment of MDR-TB with pyrazinamide irrespective of phenotypic testing, our approach suggests that structural information can and should be used to help guide patient treatment decisions.