Long-range forecasting is a notoriously difficult area that has received increased attention in recent years. Both statistical and dynamical methods have been proposed, which are all based on the assumption that the atmospheric variability is influenced on the seasonal time scale by lower boundary forcings such as the sea surface temperature (SST) or the continental hydrology. Whereas dynamical predictions rely on the explicit simulation of major atmospheric processes, it is still unclear that they lead to better estimations of seasonal precipitation anomalies on the regional scale. In the present study, both forecasting strategies are used to predict monsoon rainfall anomalies over Central Sahel. The statistical predictions are based on linear regressions using SST and rainfall predictors observed over 1968–97, a relatively dry period compared with the whole 20th century. The dynamical predictions are based on ten-member ensembles of seasonal hindcasts using the ARPEGE atmospheric model forced by observed SST over the 1979–93 period. The results suggest that the ARPEGE precipitation forecasts are less skilful than a linear regression scheme based on only two predictors. However, the atmospheric model shows a better ability to simulate the variability of the large-scale monsoon circulation. As a consequence, a simple statistical adaptation of the ARPEGE dynamical predictions is feasible and leads to better predictions of the rainfall anomalies than the direct use of the modelled precipitation. Finally, a statistico-dynamical method is proposed, based on a multivariate regression using both observed predictors and ARPEGE dynamical outputs. Although based on too short a time-series to lead to definitive conclusions, the results are encouraging and suggest the need to combine dynamical and statistical tools to perform efficient seasonal predictions. Copyright © 2002 Royal Meteorological Society.