Climate simulations require very complex numerical models. Unfortunately, they typically present biases due to parameterizations, choices of numerical schemes, and the complexity of many physical processes. Beyond improving the models themselves, a way to improve the performance of the modeled climate is to consider multi-model combinations. In the present study, we propose a method to select the models that yield a multi-model ensemble combination that efficiently reproduces target features of the observations. We used a neural classifier (self-organizing maps), associated with a multi-correspondence analysis to identify the models that best represent some target climate property. We can thereby determine an efficient multi-model ensemble. We illustrated the methodology with results focusing on the mean sea surface temperature seasonal cycle in the Senegalo- Mauritanian region. We compared 47 CMIP5 model configurations to available observations. The method allows us to identify a subset of CMIP5 models able to form an efficient multi-model ensemble. The future decrease in the Senegalo- Mauritanian upwelling proposed in recent studies is then revisited using this multi-model selection.