The Revised National Tuberculosis Control Programme (RNTCP), based on the World Health Organization's DOTS strategy,* was introduced in India in the mid-1990s. This paper reports the findings from operational research studies in two pilot sites in New Delhi from 1996 to 1998. A variety of operational research methods were used, including semi-structured interviews, focus group discussions, non-participant observations and collection of data from the tuberculosis registers. The cure rates for the clinics were 71 and 75% with a default rate of 6 and 11%, respectively. An important finding was that health workers screened patients to determine their ability to conform to the direct observation of treatment element of the RNTCP. If the health worker was confident that the patient would comply and/or be easy to trace in the community in the event of 'default', they were provided with short-course treatment under the RNTCP. Other patients, largely those who were in absolute poverty, socially marginalized, itinerant labourers, poorly integrated in the city, were put on standard tuberculosis (TB) treatment as for the previous National TB Programme. The programme was evidently excluding the most vulnerable from the best available care. These findings demonstrate the potential dangers of target-driven programmes where there is an absence of support to both frontline health workers and patients. The paper also highlights the importance of operations research in helping to identify problems within TB programmes.