Abstract The outlook for Sahel precipitation in coupled simulations of the twenty-first century is very uncertain, with different models disagreeing even on the sign of the trends. Such disagreement is especially surprising in light of the robust response of the same coupled models to the twentieth-century forcings. This study presents a statistical analysis of the preindustrial, twentieth-century and twenty-first-century A1B scenario simulations in the latest Coupled Model Intercomparison Project 3 (CMIP3) dataset; it shows that the relationship that links Sahel rainfall anomalies to tropical sea surface temperature (SST) anomalies at interannual time scales in observations is reproduced by most models, independently of the change in the basic state as the world warms. The same SST–Sahel relationship can be used to predict the simulated twentieth-century changes in Sahel rainfall from each model’s simulation of changes in Indo-Pacific SST and Atlantic SST meridional gradient, although the prediction overestimates the simulated trends. Conversely, such a relationship does not explain the rainfall trend in the twenty-first century in a majority of models. These results are consistent with there being, in most models, a substantial direct positive effect of atmospheric greenhouse gases on Sahel rainfall, not mediated through SST.