SETTING A national tuberculosis (TB) drug resistance survey in Tanzania. OBJECTIVE To compare the performance of the Genotype® MTBDRplus line-probe assay (LPA) on smear-positive sputum specimens with conventional culture and isoniazid (INH) plus rifampicin (RMP) drug susceptibility testing (DST). DESIGN Mycobacterium tuberculosis isolates tested at the Tanzanian Central TB Reference Laboratory (CTRL) were submitted for quality assurance of phenotypic DST to its supranational reference laboratory (SRL), together with ethanol-preserved sputum specimens for LPA DST. RESULTS Only 321 samples could be tested using LPA; of these, three were identified as being non-tuberculous mycobacteria using CTRL DST. Both tests had 269 sets with interpretable results. CTRL DST yielded almost the same number of interpretable results as LPA, with 90% concordance (κ = 0.612, P textless 0.001). Five (1.9%) multidrug-resistant (MDR) strains, 46 (17.1%) resistant to INH only and 0 RMP only, were found by CTRL DST. For the LPA, these results were respectively 5 (1.9%), 26 (9.7%) and 2 (0.7%). With SRL DST as the gold standard, LPA was more accurate than CTRL DST for RMP, but missed almost half the INH-resistant samples. CONCLUSION LPA applied directly on ethanol-preserved sputum specimens was similar to phenotypic DST in terms of yield of interpretable results. Although probably more accurate for RMP and MDR-TB, it appears to seriously underestimate INH resistance. Considering speed, easy and safe specimen transportation and low infrastructure requirements, LPA DST from sputum can be recommended for surveys in resource-poor settings.