This study provides statistical evidence that June--September Sahelian rainfall hindcasts currently based on oceanic thermal predictors apprehend more the negative trend than the interannual rainfall variations. Four physically meaningful predictors of June--September Sahel rainfall are first selected through the near-surface April--May information and several experimental hindcasts provided. We then discuss the skills achieved using regression techniques and cross-validated discriminant functions. In that context, 8/11 of the driest seasons and 8/10 of the wettest are correctly predicted. Finally using completely independent training and working periods we show that better and significant hindcast skills are obtained by adding regional moist static energy predictors to the large-scale sea surface temperature information. ¿ 1999 American Geophysical Union