In Sahel, scarcity and randomness of water resources are the main explanation of the variability of the plant response in multilocation trials. This variability is known for plant breeders under the name of interaction between genotype and environment (G x E). To choose a genotype for a given environment supposes to be able to predict what each genotype will produce under a range of predictable agroclimatic situations of this environment. Most of the classical G x E interaction analysis methods describe the phenotype by a function of the genotype and environment effects that can be estimated from agronomic trials. However environment is variable from year to year and this approach cannot result in good prediction of genotypes performances in a new environment. Nevertheless, factors that must be taken into account are known for a given environment. They include for each trial, rainfall, solar radiation, wind, etc. The problem is that these climatic variables are too many. Water balance models exist and make it possible to take into account the climate effects on cultures, but their parameters must be determined for each new genotype, which is expensive. We propose to predict genotypes' response in new environments taking into account their climate, without excessive costs. The tool to be developed is intended initially for Sahel plant breeders who will be able, in short plant breeding programs (2-3 years), to reduce the uncertainty of the new genotypes performances prediction. (Texte intégral)