The present study analysed the concordance among four different molecular diagnostic methods for tuberculosis (TB) in pulmonary and blood samples from immunocompromised patients. A total of 165 blood and 194 sputum samples were collected from 181 human immunodeficiency virus (HIV)-infected patients with upper respiratory complaints, regardless of suspicious for TB. The samples were submitted for smear microscopy, culture and molecular tests: a laboratory-developed conventional polymerase chain reaction (PCR) and real-time quantitative PCR (qPCR) and the Gen-Probe and Detect-TB Ampligenix kits. The samples were handled blindly by all the technicians involved, from sample processing to results analysis. For sputum, the sensitivity and specificity were 100% and 96.7% for qPCR, 81.8% and 94.5% for Gen-Probe and 100% and 66.3% for Detect-TB, respectively. qPCR presented the best concordance with sputum culture [kappa (k) = 0.864)], followed by Gen-Probe (k = 0.682). For blood samples, qPCR showed 100% sensitivity and 92.3% specificity, with a substantial correlation with sputum culture (k = 0.754) and with the qPCR results obtained from sputum of the corresponding patient (k = 0.630). Conventional PCR demonstrated the worst results for sputa and blood, with a sensitivity of 100% vs. 88.9% and a specificity of 46.3% vs. 32%, respectively. Commercial or laboratory-developed molecular assays can overcome the difficulties in the diagnosis of TB in paucibacillary patients using conventional methods available in most laboratories.