Background: Tuberculosis continues to be a global health problem. Objective: To evaluate an automated system designed to diagnose tuberculosis, comparing it to sputum microscopy and culture in Löwenstein-Jensen medium. Method: A comparative study using 844 sputum samples, collected between September and December of 1999 at a reference center for tuberculosis in São Paulo, Brazil, to draw distinctions between the results obtained through the use of the automated system and those obtained through sputum microscopy and culture in Löwenstein-Jensen medium. Results: Of the 844 samples evaluated, 27.1% tested positive for acid-fast bacilli, and 72.9% tested negative. In Löwenstein-Jensen culture, 34.7% were positive and 63% were negative, compared with 37.1% positivity and 56.9% negativity using the automated system. Sensitivity was 98.1% for the automated system and 91.9% for Löwenstein-Jensen culture. Specificity and positive predictive value were 100% for both methods. Negative predictive value was 98.9% for the automated system and 95.5% for Löwenstein-Jensen culture. The degree of accuracy was 99.3% for the automated system and 97% for Löwenstein-Jensen culture, and the Kappa was 0.99 for the automated system and 0.94 for Löwenstein-Jensen culture. The difference between the mean time to detection of mycobacteria using the automated system (10.5 days) and that found using Löwenstein-Jensen culture (34.7 days) was statistically significant. Conclusion: The difference between the culture yield obtained using the automated system and that achieved with Löwenstein-Jensen culture was statistically significant. Mean time to detection of mycobacteria was significantly shorter with the automated system. The higher yield provided by this new system justifies its use in a reference center for tuberculosis in São Paulo

Keywords: Mycobacterium tuberculosis. Diagnostic. Automated systems.