Rapid and sensitive detection of mycobacteria is essential for the successful treatment and control of mycobacterial infections. It is considered good laboratory practice to combine a liquid medium with at least one solid medium for culturing mycobacteria from clinical specimens. Automated systems with liquid media provide faster results and are more sensitive, while solid media help to detect mixed infections and can detect mycobacteria that do not grow in the broth medium. In order to evaluate the BacT/ALERT 3D system (bioMérieux, Durham, NC, USA) and to find the optimal combination of media, we compared the results achieved with BacT/ALERT 3D, Lowenstein-Jensen (L-J) medium (Bio-Rad, Marne-la Coquette, France), and the Middlebrook 7H10/7H11 biplate (Becton Dickinson, Sparks, MD, USA) on consecutive clinical specimens collected within a period of 26 months. A total of 6,250 clinical samples (4,051 respiratory specimens, 1,209 normally sterile body fluids, 292 biopsy specimens, 247 blood specimens, 344 urine specimens and 304 other specimens) from 3,247 patients were tested for the presence of mycobacteria. Specimens with a different panel of media (blood and urine) and specimens contaminated with bacteria other than mycobacteria were excluded from the evaluation. Tissue specimens were crushed and homogenized. Homogenized tissue, respiratory specimens and other specimens from contaminated sites were decontaminated with sodium dodecyl sulfate-NaOH as described by Tacquet and Tisson [1]. Liquids from sterile sites were concentrated by centrifugation (3000 g, 15 min). Aliquots of 0.3-0.5 ml were inoculated onto a L-J slant, a Middlebrook 7H10/7H11 selective agar biplate with carbenicillin, polymyxin B, trimethoprim lactate, and amphotericin B, and into a BacT/ALERT MP bottle. For specimens from