This work aims to evaluate the performance of outputs in six regional climate models (RCMs) of the Cordex-Africa program (RCA4, CRCM5, RACMO22T, HIRHAM5, REMO2009, CCLM4-8-17), at the Medjerda Watershed. It is based on the evolution of precipitation during the reference period (1981-2005) using different evaluation criteria (Pearson’s correlation coefficient (r), the root mean square error (RMSE), the percentage of bias (PBIAS), the Taylor Diagram and Taylor’s Skill Score (TSS)). The results confirm the difference between the RCMs in the estimation of climatic variables at different time scales (annual, seasonal and monthly). The annual precipitation regime has been accurately predicted by REMO2009, which presents a result of 0% in terms of PBIAS between observations and simulations. The agreement between the observed and simulated precipitations at a seasonal time scale is superior in terms of the correlation r = 1 by REMO2009, RACMO22T, CCLM4-8-17, CRCM5, HIRHAM5. In terms of RMSE the REMO2009 model scored a low value of 1.13 mm in autumn, on a monthly time scale, while the CCLM4- 8-17 model showed a high performance value in terms of Taylor’s Skill Score (TSS = 0.91). In overall terms, those models, which reproduce the observed climatology of the Medjerda basin more efficiently, and which may be considered the most efficient, include REMO2009, CCLM4-8-17, RACMO22T and CRCM5. This study recommends the use of these selected models for future precipitation projections in the Medjerda Watershed and it is therefore necessary to correct the outputs of the regional models of the Cordex-Africa project and to use the set of those models that perform most efficiently with respect to future climate change impacts and for adaptation studies on the African continent.