Previous studies on the effect of breastfeeding on mother-to-child transmission (MTCT) of HIV have not attempted to examine the influence of different types of breastfeeding practice. To attempt to address some of these inadequacies, infant feeding practices of 549 HIV-infected women involved in a trial in Durban, South Africa were documented prospectively. Women were counseled on infant feeding choices according to UNAIDS guidelines, and those who chose to breastfeed were encouraged to practice exclusive breastfeeding. The MTCT rates of HIV-1 at 3 months were compared in the three different feeding groups (never breastfed, exclusive breastfeeding, and mixed breastfeeding). At 3 months, 18.8% of 156 never-breastfed children were infected compared to 21.3% of 393 breastfed children (p = 0.50). Children exclusively breastfed to at least 3 months (n = 103) were less likely to be infected (14.3%) than those receiving mixed feeding before 3 months (24.1%) (p = 0.03). After adjustment for potential confounders (maternal CD4:CD8 ratio, syphilis screening test results, and preterm delivery), exclusive breastfeeding carried a significantly lower risk of HIV-1 transmission than mixed feeding (hazard ratio [HR] 0.52, 95% CI: 0.28-0.98) and an equivalent risk to no breastfeeding (HR 0.85, 95% CI: 0.51-1.42). Our findings have important implications for HIV and infant feeding policies in developing countries, and it is critical that further research be undertaken. In the meantime, breastfeeding policies for HIV-infected women require urgent review. If confirmed, exclusive breastfeeding may offer HIV-infected women in developing countries an affordable, culturally acceptable, and effective means of reducing MTCT of HIV-1 while maintaining the overwhelming benefits of breastfeeding.