OBJECTIVE: Virulent strains of the Mycobacterium tuberculosis complex, under certain appropriate conditions, grow as characteristic ropes, bundles or serpentine cords known as cord factor or growth in cords. The objective of the present study was to evaluate cord factor detection as a method of achieving presumptive identification of the M. tuberculosis complex, comparing it to conventional typing tests. METHODS: A total of 743 strains were analyzed from January of 2002 to December of 2005 in the Mycobacteria Sector of the Adolfo Lutz Institute, located in the city of Santos, Brazil. Samples were obtained from clinical specimens collected from patients with respiratory symptoms treated at basic health clinics in the greater metropolitan area of Santos. Ziehl-Neelsen-stained smears were prepared, 301 (40.5%) in MB/BacT broth and 442 (59.5%) on solid media, either Lowenstein-Jensen or Ogawa-Kudoh. RESULTS: The sensitivity, specificity, positive predictive value and negative predictive value obtained during the performance comparison of the two methods (cord factor detection and conventional typing) using both isolation media were, respectively, 98.5, 88, 97 and 93%. The method was more sensitive on solid medium (100%), and the difference in sensitivity between the two media types was only 2.7%. CONCLUSIONS: Taking into consideration the results obtained, we conclude that, in laboratories with a high incidence of M. tuberculosis complex isolation and limited economic resources, cord factor detection is a fast and valid criterion for identifying these mycobacteria using liquid or solid medium. It also enables subsequent conclusive identification tests, as well as additional sensitivity tests when necessary.