Abstract. Snow processes are a key component of the water cycle in mountainous areas as well as in many areas of the mid- and high latitudes of the Earth. The complexity of these processes, coupled with the limited data available on them, has led to the development of different modelling approaches to improve their understanding and support management practices. Physically-based approaches, such as the energy balance method, provide the best representation of these processes but at the expense of high data requirements. Data limitations, in most situations, constrain the use of these methods in favour of more simple approaches. The temperature-index method is the most widely-used modelling approach of the snowpack processes for hydrological modelling, with many variants implemented in different models. However, in many cases, the decisions on the most suitable complexity of these conceptualisations are not adequately assessed for a given model structure, application, or decision-making support tool. In this study, we assessed model structure choices of the HBV model, a popular semi-distributed, bucket-type hydrological model, for its application in mountainous areas in Central Europe. To this end, we reviewed the most widely-used choices to different components of the snow routine in different hydrological models and proposed a series of modifications to the structure of HBV. We constrained the choice of modifications to those that are aligned with HBV’s modelling approach of keeping processes as simple as possible to constrain model complexity. We analysed a total of 64 alternative snow routine structures over 54 catchments using a split-sample test. We found that using (a) an exponential snowmelt function coupled with no refreezing instead of a linear function for both processes and (b) a seasonally-variable degree-day factor instead of a constant one were, overall, the most valuable modifications to the model. Additionally, we found that increasing model complexity does not necessarily lead to improved model performance per se. Instead, we found that a thorough analysis of the different processes included in the model and their optimal degree of realism for a given application is a preferable alternative. While the results may not be transferrable to other modelling purposes or geographical domains, the methodology presented here may be used to assess the suitability of model design choices.