Problem solving has been considered the gold standard of mathematical activity. It is a goal of mathematics education that students become problem solvers, and it is suggested that problem solving is a superior method for learning mathematics. However, the arguments supporting the claim that problem solving leads to better learning are often vague. In specific studies, problem solving often constitutes mere one part of a compound design, making it difficult to determine the specific contribution of problem solving. The aim of this paper is to develop an initial theoretical model for problem solving as a learning activity, based on existing frameworks and previous research. Suggestions for how this model could be empirically tested are also discussed.