OBJECTIVE To test the hypothesis that subclinical Mycobacterium avium intracellulare complex (MAC) infection may result in the development of a tuberculin response in immunodeficient HIV-infected individuals treated with zidovudine. METHODS Longitudinal, observational study. METHODS The Western Australian HIV Cohort Study; a prospective, single centre, population-based observational study of the natural history of HIV disease. METHODS Forty-nine patients with impaired delayed-type hypersensitivity (DTH) responses and negative tuberculin responses in whom DTH responses were augmented within 6 months of starting zidovudine therapy. METHODS Progression to disseminated MAC infection stratified according to the presence or absence of a tuberculin response in the first 6 months of zidovudine therapy. RESULTS Twenty-nine of the patients developed a post-zidovudine tuberculin response. None of the tuberculin non-responders developed disseminated MAC infection during the study period; the Kaplan-Meier probability estimate of disseminated MAC infection was 50% at 24 months and reached 100% 40 months after zidovudine was commenced in tuberculin responders. All patients with disseminated MAC infection had become anergic to all antigens, including tuberculin, before diagnosis. The probability of a post-zidovudine tuberculin response was related to the severity of peripheral blood CD4+ T-cell depletion, rising from an estimated 20% at 20% CD4+ T cells to 100% at < or = 1% CD4+ T cells. CONCLUSIONS The restoration of a cellular immune response against subclinical MAC infection can be demonstrated by measuring the DTH response to tuberculin in patients with impaired DTH augmented by zidovudine therapy. The findings suggest that MAC infection is almost inevitable, but often asymptomatic, in profoundly immunodeficient HIV-infected patients and that a prolonged subclinical phase of MAC infection is usual.