In the framework of the ENSEMBLES FP6 project, an ensemble prediction system based on five different state-of-the-art European coupled models has been developed. This study evaluates the performance of these models for forecasting the West African monsoon (WAM) at the monthly time scale. From simulations started the 1 May of each year and covering the period 1991–2001, the reproducibility and potential predictability (PP) of key parameters of the WAM—rainfall, zonal and meridional wind at four levels from the surface to 200 hPa, and specific humidity, from July to September—are assessed. The Sahelian rainfall mode of variability is not accurately reproduced contrary to the Guinean rainfall one: the correlation between observations (from CMAP) and the multi-model ensemble mean is 0.17 and 0.55, respectively. For the Sahelian mode, the correlation is consistent with a low PP of about ~6%. The PP of the Guinean mode is higher, ~44% suggesting a stronger forcing of the sea surface temperature on rainfall variability over this region. Parameters relative to the atmospheric dynamics are on average much more skillful and reproducible than rainfall. Among them, the first mode of variability of the zonal wind at 200 hPa that depicts the Tropical Easterly Jet, is correlated at 0.79 with its “observed” counterpart (from the NCEP/DOE2 reanalyses) and has a PP of 39%. Moreover, models reproduce the correlations between all the atmospheric dynamics parameters and the Sahelian rainfall in a satisfactory way. In that context, a statistical adaptation of the atmospheric dynamic forecasts, using a linear regression model with the leading principal components of the atmospheric dynamical parameters studied, leads to moderate receiver operating characteristic area under the curve and correlation skill scores for the Sahelian rainfall. These scores are however much higher than those obtained using the modelled rainfall.