Human immunodeficiency virus type 1 (HIV-1) is transmitted mainly by cell-to-cell contact. We postulated that transmission of HIV-1 through breastmilk could be favoured by the presence of infected cells, by deficiency of anti-infective substances in breastmilk, or both factors. 215 HIV-1-infected women were enrolled at delivery in Kigali, Rwanda; milk samples were collected 15 days, 6 months, and 18 months post partum. HIV-1 IgG, secretory IgA, and IgM were assayed by western blot, for the latter two after removal of IgG with protein G. In the 15-day and 6-month samples, we sought viral genome in milk cells by a double polymerase chain reaction with three sets of primers (gag, pol, and env). HIV-1 infection in the offspring was defined according to serological and clinical criteria. At 15 days, 6 months, and 18 months post partum, HIV-1 specific IgG was detected in 95%, 98%, and 97% of breastmilk samples, IgA in 23%, 28%, and 41%, and IgM in 66%, 78%, and 41%. In children who survived longer than 18 months, the probability of infection was associated with lack of persistence of IgM and IgA in their mothers' milk (adjusted chi 2 for trend, p = 0.01 for IgM and p = 0.05 for IgA). The presence of HIV-1-infected cells in the milk 15 days post partum was strongly predictive of HIV-1 infection in the child, by both univariate (p < 0.05) and multivariate analysis (p = 0.01). The combination of HIV-1-infected cells in breastmilk and a defective IgM response was the strongest predictor of infection. HIV-1 infection in breastfed children born to infected mothers is associated with the presence of integrated viral DNA in the mothers' milk cells. IgM and IgA anti-HIV-1 in breastmilk may protect against postnatal transmission of the virus.