Serum immunoglobulin (Ig) G responses to a variety of mycobacterial antigens were measured in children from the UK, in children with tuberculosis from Hyderabad, India and Dhaka, Bangladesh, classified according to whether the disease was disseminated or localized, and in non-tuberculous controls. Anti-lipoarabinomannan (LAM) IgG responses in UK children showed a marked trough between 6 months and 3 years coincident with the reported peak incidence of disseminated tuberculosis. Geometric mean IgG responses to sonicates of slow-growing mycobacteria (rich in LAM) in 36 children with disseminated tuberculosis were markedly lower than in 99 children with localized tuberculous lesions (for Mycobacterium scrofulaceum P < 0.01, for M. tuberculosis P < 0.01, and for M. vaccae P < 0.01). Responses to purified LAM were also lower in the disseminated tuberculosis group (P < 0.05) but there was no difference between the groups in their response to mycobacterial 65 kDa protein. Multiple regression analysis showed that the reduced response to sonicated mycobacterial antigens and to LAM in children with disseminated disease was independent of age, nutritional status, skin test reactivity, duration of previous symptoms, and city of origin. There was no evidence for sequestration of antibody to immune complexes. These findings are compatible with the hypothesis that children with low levels of antibody to sonicated mycobacterial antigen and to LAM, or those who cannot mount an antibody response, are predisposed to dissemination. A role for antibody in preventing disseminated forms of tuberculosis in childhood has implications for the development of improved vaccines and for the optimum timing of vaccination with bacille Calmette-Guérin.