This PhD thesis presents the development of a methodological framework to analyse potential climate change impacts on a high mountainous water resources system and to quantify the associated modelling uncertainties. The main objective is to show whether state-of-the-art hydrological modelling techniques driven by currently available climate change scenarios enable a prediction of the long-term evolution of the analysed system. The case study is a highly glacierized catchment feeding a hydropower plant located in the Swiss Alps. The climate change impact analysis is based on a classical simulation approach: The system behaviour is modelled for an observed control period (1961 to 1990) and for a future period (2070 to 2099) characterised by a modified (predicted) climate. The climate change impact on the studied system is assessed through the comparison of some key characteristics of the system for the two periods (e.g. the mean annual discharge or the hydropower production). The system simulation is completed through a set of four models, a water management model, a hydrological discharge model, a glacier surface evolution model and a model for the production of local scale meteorological time series (precipitation and temperature) based on global and regional climate model outputs. The local scale models have been specifically developed for the purposes of this thesis. For each of them, an appropriate statistical method for the quantification of the inherent modelling uncertainties has been developed. A special emphasis is given to the modelling uncertainties induced by the conceptual hydrological model. A method has been developed to quantify the statistical and the multi-objective modelling uncertainty in a multi-model framework including several equivalent model structures. This method has been specially designed for the quantification of the prediction uncertainty in climate change impact studies but it is transposable to other hydrological modelling contexts. The overall prediction uncertainty and the contribution of each source of modelling uncertainty is quantified through Monte Carlo simulations of the system behaviour combining successively the different sources of modelling uncertainty. It is shown that the uncertainties induced by the prediction of the climate evolution are much higher than the ones induced by the local scale models of the system behaviour. The uncertainty related to the use of different regional climate models is however nearly as important as the one due to the choice of the underlying global climate model and the green house gas emission scenario. Using a fixed hydrological model structure, the predicted climate evolution induces a significant reduction of the hydropower production performance due to a considerable hydrological regime modification. The available data, the current discharge modelling techniques and the knowledge about the underlying processes are however not sufficient to chose an objectively best model structure. Considering the multi-model approach (including different hydrological model structures), an unambiguous prediction of the hydrological reaction to the analysed climate change reveals impossible at the given temporal horizon.