Abstract Multidrug-resistant tuberculosis (MDRTB) is a serious world health problem that limits public actions to control tuberculosis, because the most used anti-tuberculosis first-line drugs fail to stop mycobacterium spread. Consequently, a quick detection through molecular diagnosis is essential to reduce morbidity and medical costs. Despite the availability of several molecular-based commercial-kits to diagnose multidrug-resistant tuberculosis, their diagnostic value might diverge worldwide since Mycobacterium tuberculosis genetic variability differs according to geographic location. Here, we studied the predictive value of four common mycobacterial mutations in strains isolated from endemic areas of Brazil. Mutations were found at the frequency of 41.9% for katG, 25.6% for inhA, and 69.8% for rpoB genes in multidrug-resistant strains. Multimarker analysis revealed that combination of only two mutations (“katG/S315T + rpoB/S531L”) was a better surrogate of multidrug-resistant tuberculosis than single-marker analysis (86% sensitivity vs. 62.8%). Prediction of multidrug-resistant tuberculosis was not improved by adding a third or fourth mutation in the model. Therefore, rather than using diagnostic kits detecting several mutations, we propose a simple dual-marker panel to detect multidrug-resistant tuberculosis, with 86% sensitivity and 100% specificity. In conclusion, this approach (previous genetic study + analysis of only prevalent markers) would considerably decrease the processing costs while retaining diagnostic accuracy.