HIV contributes substantially to child mortality, but factors underlying these deaths are inadequately described. With individual data from seven randomised mother-to-child transmission (MTCT) intervention trials, we estimate mortality in African children born to HIV-infected mothers and analyse selected risk factors. Early HIV infection was defined as a positive HIV-PCR test before 4 weeks of age; and late infection by a negative PCR test at or after 4 weeks of age, followed by a positive test. Mortality rate was expressed per 1000 child-years. We investigated the effect of maternal health, infant HIV infection, feeding practices, and age at acquisition of infection on mortality assessed with Cox proportional hazards models, and allowed for random effects for trials grouped geographically. 378 (11%) of 3468 children died. By age 1 year, an estimated 35.2% infected and 4.9% uninfected children will have died; by 2 years of age, 52.5% and 7.6% will have died, respectively. Mortality varied by geographical region, and was associated with maternal death (adjusted odds ratio 2.27, 95% CI 1.62-3.19), CD4+ cell counts <200 per microL (1.91, 1.39-2.62), and infant HIV infection (8.16, 6.43-10.33). Mortality was not associated with either ever breastfeeding and never breastfeeding in either infected or uninfected children. In infected children, mortality was significantly lower for those with late infection than those with early infection (0.52, 0.39-0.70). This effect was also seen in analyses of survival from the age at infection (0.74, 0.55-0.99). These findings highlight the necessity for timely antiretroviral care, for support for HIV-infected women and children in developing countries, and for assessment of prophylactic programmes to prevent MTCT, including child mortality and infection averted.