Much remains unknown about latent infection with Mycobacterium tuberculosis. Existing immunodiagnostic tools for this condition have various limitations, most importantly in their ability to predict disease. Randomised controlled trials have established protective efficacy of isoniazid therapy for 6-12 months among non-HIV-infected and HIV-infected subjects. While efficacy may reach 90%, acceptance and adherence to prolonged therapy are less than desired. Rifampicin plus pyrazinamide for 2 months, though efficacious, has been associated with excess hepatotoxicity in non-HIV-infected persons. Isoniazid plus rifampicin for 3 months has proven efficacy, but adverse effects may be more frequent than isoniazid or rifampicin monotherapy. Rifampicin monotherapy for 3-4 months is well tolerated, but efficacy data are currently limited, and concerns remain over possible selection of rifampicin-resistant mutants. For contacts of patients with multidrug-resistant tuberculosis, expert opinions differ on whether to treat with at least two drugs or just a fluoroquinolone, and for how long. With the existing diagnostic and treatment tools, efficacy of preventive therapy does not necessarily translate into field effectiveness. A targeted approach is required to maximise cost-effectiveness. Each geographic region needs to set its own priority after taking into account available scientific data and local circumstances.