Traditionally, snowmelt modelling has been governed by the operational need for runoff forecasts. Parsimony in terms of model complexity and data requirements was a major concern. More recently, the increased importance of analyzing environmental problems and extreme conditions has motivated the development of distributed snow models. Unfortunately, the use of this type of models is limited by a number of factors including a) the extreme heterogeneity of the hydrologic environment, b) the mismatch of scales between observed variables and model state variables, c) the large number of model parameters, and d) the observability/testability problem. This paper discusses the implications of these constraints on the use of site and catchment scale concepts, regionalisation techniques, and calibration methods. In particular, the point is made that in many cases model parameters are poorly defined or not unique when being optimized on the basis of runoff data. Snow cover depletion patterns are shown to be vastly superior to runoff data for discriminating between alternative model assumptions. The patterns are capable of addressing individual model components representing snow deposition and albedo while the respective parameters are highly intercorrelated in terms of catchment runoff. The paper concludes that site scale models of snow cover processes are fairly advanced but much is left to be done at the catchment scale. Specifically, more emphasis needs to be directed towards measuring and representing spatial variability in catchments as well as on spatially distributed model evaluation.