Drug susceptibility testing (DST) is often needed in patients clinically failing tuberculosis (TB) therapy. Most studies of phenotypic direct drug susceptibility tests, such as microscopic observation drug susceptibility (MODS) tests, have been performed in patients not receiving TB treatment. The effect of ongoing TB treatment on the performance of MODS direct DST has not been previously explored, but patients failing such therapy constitute an important target group. The aim of this study was to determine the performance of MODS direct rifampicin and isoniazid DST in patients clinically failing first-line TB treatment, and to compare MODS direct DST with indirect proportion method DST. Sputa from 264 TB patients were cultured in parallel in Lowenstein-Jensen (LJ) and MODS assays; strains were tested for rifampicin and isoniazid susceptibility by the proportion method at the national reference laboratory. Ninety-three samples were culture-positive by LJ and MODS (concordance of 96%; kappa 0.92). With conventional MODS plate DST reading (performed on the same day as the sample is classified as culture-positive), the isoniazid DST concordance was 96.8% (kappa 0.89), and the concordance for rifampicin susceptibility testing was 92.6% (kappa 0.80). Reading of MODS DST plates 1 week after cultures had been determined to be culture-positive improved overall performance marginally-the isoniazid DST concordance was 95.7% (kappa 0.85); and the rifampicin DST concordance was 96.8% (kappa 0.91). Sensitivity for detection of multidrug-resistant TB was 95.8%. MODS testing provided reliable rifampicin and isoniazid DST results for samples obtained from patients receiving TB therapy. A modified DST reading schedule for such samples, with a final reading 1 week after a MODS culture turns positive, marginally improves the concordance with reference DST.