Existing studies of the quality of tuberculosis care have relied on recall-based patient surveys, questionnaire surveys of knowledge, and prescription or medical record analysis, and the results mostly show the health-care provider's knowledge rather than actual practice. No study has used standardised patients to assess clinical practice. Therefore we aimed to assess quality of care for tuberculosis using such patients. We did a pilot, cross-sectional validation study of a convenience sample of consenting private health-care providers in low-income and middle-income areas of Delhi, India. We recruited standardised patients in apparently good health from the local community to present four cases (two of presumed tuberculosis and one each of confirmed tuberculosis and suspected multidrug-resistant tuberculosis) to a randomly allocated health-care provider. The key objective was to validate the standardised-patient method using three criteria: negligible risk and ability to avoid adverse events for providers and standardised patients, low detection rates of standardised patients by providers, and data accuracy across standardised patients and audio verification of standardised-patient recall. We also used medical vignettes to assess providers' knowledge of presumed tuberculosis. Correct case management was benchmarked using Standards for Tuberculosis Care in India (STCI). Between Feb 2, and March 28, 2014, we recruited and trained 17 standardised patients who had 250 interactions with 100 health-care providers, 29 of whom were qualified in allopathic medicine (ie, they had a Bachelor of Medicine & Surgery [MBBS] degree), 40 of whom practised alternative medicine, and 31 of whom were informal health-care providers with few or no qualifications. The interactions took place between April 1, and April 23, 2014. The proportion of detected standardised patients was low (11 [5%] detected out of 232 interactions among providers who completed the follow-up survey), and standardised patients' recall correlated highly with audio recordings (r=0·63 [95% CI 0·53-0·79]), with no safety concerns reported. The mean consultation length was 6 min (95% CI 5·5-6·6) with a mean of 6·18 (5·72-6·64) questions or examinations completed, representing 35% (33-38) of essential checklist items. Across all cases, only 52 (21% [16-26]) of 250 were correctly managed. Correct management was higher among MBBS-qualified doctors than other types of health-care provider (adjusted odds ratio 2·41 [95% CI 1·17-4·93]; p=0·0166). Of the 69 providers who completed the vignette, knowledge in the vignettes was more consistent with STCI than their actual clinical practice-eg, 50 (73%) ordered a chest radiograph or sputum test during the vignette compared with seven (10%) during the standardised-patient interaction; OR 0·04 (95% CI 0·02-0·11); p<0·0001. Standardised patients can be successfully implemented to assess tuberculosis care. Our data suggest a big gap between private provider knowledge and practice. Additional work is needed to substantiate our pilot data, understand the know-do gap in provider behaviour, and to identify the best approach to measure and improve the quality of tuberculosis care in India. Grand Challenges Canada, the Bill & Melinda Gates Foundation, Knowledge for Change Program, and the World Bank Development Research Group.