Understanding why different climate models produce different projections of future regional climate change is critical for the evaluation of the reliability of their projections. This study investigates the causes of projection spread across CMIP5 models for seasonal mean rainfall change over East Africa. The methodological approaches could also be applied to rainfall change in other parts of Africa. One outcome, building on earlier studies, is that differences amongst models’ responses to anthropogenic emissions can be caused by mechanisms that differ from those that cause natural year-to-year variations. This implies that simple measures of model performance based on traditional metrics of their current climate simulation, may not always be relevant to their performance for longer term future change. Specific findings for the Short Rains show that the CMIP5 spread in projected rainfall change is primarily due to uncertain responses to the slow warming of the oceans (but not uncertainty in the warming itself), and a small direct impact of rising CO2 concentrations. These in turn are due to uncertain regional dynamics, both over Africa and over remote regions, rather than globally-coherent (thermo)dynamics. For the Long Rains, results are similar, except that some of the key regional uncertainties are very clearly located beyond Africa. This is found to reflect the behaviour of two versions of an outlying model which experience exceptional warming in the southern subtropical oceans, from which large lower tropospheric moisture anomalies are advected by the mean flow to contribute to exceptional increases in the Long Rains totals. Further research could lead to a useful assessment of the reliability of these exceptional projections.