High rates of infant mortality in Africa continue to be a major public health concern today, despite the fact that most deaths can be prevented from well known, relatively low cost technologies. Using multiple years of DHS from four countries, we estimate the change in the relative risk of death as well as the main contributions to the change in mortality over time. We find significant declines in the mortality hazard in each of the 4 countries, with the largest declines in Malawi (44 percent) and Tanzania (22 percent) between the mid 1990s to mid 2000s, although there is significant variation by age group in the hazard rate across time. In Zambia for example, the hazard increased for children ages 25-60 months in spite of on overall decline in mortality, while in Mozambique the largest decline in mortality was exactly among this age group. The decomposition analysis illustrates that some of the main correlates of mortality did not contribute to overall declines over time, because the levels of these correlates did not change during the study period. This is particularly true for birth spacing, attended births and breastfeeding. The analysis also demonstrates the overall lack of explanatory power of the individual and household level variables available for use in the DHS, indicating the need to collect complementary supply side information, through community questionnaires for example, that can be linked to DHS households and thus expand the set of covariates available for modeling child survival and other health outcomes.