With current concerns of the global threat from tuberculosis, it has become important to rapidly identify the bacteria. Traditional technologies involving isolation and amplification of the pathogenic bacteria are complicated and time-consuming. In this work, we describe a sensitive NMR-based detection method to identify bacteria via bacteria-induced self-assembly of magnetic nanoparticles. Bacillus Calmette-Guérin (BCG) was used as a surrogate for Mycobacterium tuberculosis. We prepared the probes by covalent immobilization of the anti-BCG monoclonal antibody onto carboxylic acid-functionalized magnetic nanoparticles. Once a solution containing BCG was introduced, the probes switched from a well dispersed state to an aggregated one, leading to a distinct and dose-dependent change in the spin–spin relaxation time (T2) of the solution. Thus the qualitative and quantitative detection method for BCG was established. The method provides specific detection of as few as 8 bacterial cells per millilitre in experimental samples within 1 hour, which will be promising for rapid detection ofM. tuberculosis in clinical samples.