Malaria can, a priori, kill on its own ('direct' malaria mortality) or in conjunction with some other cause(s) ('indirect' malaria mortality). There are three distinct approaches to the measurement of malaria mortality: (1) measurement of malaria-specific mortality (based on attributing each death to a single cause) or its surrogate, the admission rate for life-threatening malaria; (2) measurement of the reduction in all-cause mortality after removal (or near removal) of malaria; and (3) the estimation of the malaria mortality required to explain the observed frequency of the HbS gene. There is a strong indication that approaches (2) and (3) yield estimates of total (direct and indirect) malaria mortality which are at least twice as high as those obtained using approach (1), which probably measures mostly direct malaria mortality. There is currently a controversy about the long-term impact on mortality of reducing the intensity of malaria transmission from 'high' to 'intermediate', given the expected loss of immunity. Certain geographical comparisons (of the results of 'nature's experiment') indicate that, when the intensity of transmission is high, the incidence of life-threatening malaria falls. However, the seasonal variation in the intensity of transmission decreases with increases in the intensity; low seasonal variation is probably beneficial, allowing a smoother transition from passive to active immunity, but is unlikely to be reversed by preventive measures. It also seems likely that geographical comparisons produce estimates of predominantly direct mortality, which might not run in parallel with the indirect mortality. A tentative numerical exploration of the problem, using a simple demographic model, indicates that, if indirect malaria mortality is important and the other causes of death are concentrated in early life, the long-term impact on all-cause mortality of reducing exposure (although less than the short-term) will probably always be beneficial, even if there is some increase in direct malaria mortality.