The most common method for detection of drug resistant (DR) TB in resource-limited settings (RLSs) is indirect susceptibility testing on Lowenstein-Jensen medium (LJ) which is very time consuming with results available only after 2-3 months. Effective therapy of DR TB is therefore markedly delayed and patients can transmit resistant strains. Rapid and accurate tests suitable for RLSs in the diagnosis of DR TB are thus highly needed. In this study we compared two direct techniques--Nitrate Reductase Assay (NRA) and Microscopic Observation Drug Susceptibility (MODS) for rapid detection of MDR-TB in a high burden RLS. The sensitivity, specificity, and proportion of interpretable results were studied. Smear positive sputum was collected from 245 consecutive re-treatment TB patients attending a TB clinic in Kampala, Uganda. Samples were processed at the national reference laboratory and tested for susceptibility to rifampicin and isoniazid with direct NRA, direct MODS and the indirect LJ proportion method as reference. A total of 229 specimens were confirmed as M. tuberculosis, of these interpretable results were obtained in 217 (95%) with either the NRA or MODS. Sensitivity, specificity and kappa agreement for MDR-TB diagnosis was 97%, 98% and 0.93 with the NRA; and 87%, 95% and 0.78 with the MODS, respectively. The median time to results was 10, 7 and 64 days with NRA, MODS and the reference technique, respectively. The cost of laboratory supplies per sample was low, around 5 USD, for the rapid tests. The direct NRA and MODS offered rapid detection of resistance almost eight weeks earlier than with the reference method. In the study settings, the direct NRA was highly sensitive and specific. We consider it to have a strong potential for timely detection of MDR-TB in RLS.