Numerical models of snow avalanches propagation have acquired a central role among decision tools for avalanche protection engineering. Nevertheless, the systematic implementation of these models still faces a number of difficulties including the precise evaluation of avalanche release depths. In our work, the predetermination of snow depths in any potential release zone is achieved by spatial interpolation of the daily precipitation data acquired in 40 meteorological sites in the French Alps since 1966. Given the rarity of avalanches, extreme precipitation data (annual maxima) are considered. We use the formal framework of max-stable processes which are the generalization of univariate extreme value theory to the spatial multivariate case. Using different models for the spatial evolution of the parameters of the GEV distribution, we are able to establish snow precipitation maps for different return periods. The results show that, for a given return period and at fixed elevation, snowfalls are higher in the NE of the French Alps. However variance maxima are located in the SE which corresponds to the Mediterranean influence that tends to bring more variability. Finally, we show that the spatial dependence of extreme snow precipitations depends on the orientation of the local Alpin axis. Sensitivity of the model to the extremal dependence and to the spatial evolution of the GEV parameters is discussed. Cross-validation is used to demonstrate the robustness of the retained model.