Snow is a very active photochemical reactor that considerably affects the composition and chemistry of the lower troposphere in polar regions. Snow photochemistry models have therefore been recently developed to describe these processes. In all those models, the chemically active medium is a brine formed at the surface of snow crystals by impurities whose presence cause surface melting. Reaction and photolysis rate coefficients are those measured in dilute liquid solutions. Here, we critically examine the basis for these models by considering the structure of ice crystal surfaces, the processes involved in the interactions between impurities and ice crystals, the location of impurities in snow, and the reactivity of impurities in the various media present in snow. We conclude that the brine formed by impurities can only be present in grooves at grain boundaries and cannot cover ice crystal surfaces because of insufficient ice wettability. It is then very likely that most reactions in snow do not take place in liquids, but rather either on an actual ice surface highly different from a liquid or in particulate matter contained in snow, such as organic particles that are thought to contain most snow chromophores. We discuss why some snow models appear to adequately reproduce some observations, concluding that they are insufficiently constrained and that the use of adjustable parameters allows acceptable fits. We discuss the complexity of developing a snow model without adjustable parameters and with a predictive value. We conclude that reaching this goal in the near future is a tremendous challenge. Modeling attempts focused on snow where the impact of organic particles is minimal, such as on the east Antarctic plateau, represents the best chance of midterm success.