Reduction of deaths associated with malaria in children is a primary goal of malaria control programmes in Africa, but there has been little discussion about how changes in mortality will be measured. This paper assesses recent historical changes in the contribution of malaria to child survival in Africa by examining data from demographic surveillance systems (DSS) in 25 mainly rural settings. The data were adjusted for the varying sensitivity and specificity of verbal autopsies (VA) in different ranges of malaria mortality and for varying parasite prevalences. Average malaria mortality in the DSS sites in west Africa was 7.8 per 1000 child-years between 1982 and 1998; the rate did not change significantly over this period. In the sites in east and southern Africa combined, malaria mortality was 6.5 per 1000 child-years between 1982 and 1989, but it increased to 11.9 per 1000 child-years between 1990 and 1998. All-cause child mortality and non-malaria mortality, by contrast, decreased significantly over time in both regions; consequently, the proportion of deaths due to malaria rose from 18% to 23% in west African sites and from 18% to 37% in east and southern African sites between 1982-89 and 1990-98. If malaria mortality fell at a rate consistent with the Roll Back Malaria target of halving malaria mortality by the year 2010, an individual DSS of a total population of 63 500 could with adequate VA adjustment detect this reduction after 7 years.