Empirical studies of business process modeling typically aim at understanding factors that can improve model quality. We identify two limitations of such studies. First, the quality dimensions usually addressed are mainly syntactic and pragmatic, not addressing semantic quality sufficiently. Second, while findings related to model understanding have been anchored in cognitive theories, findings related to model construction have remained mostly unexplained. This paper proposes to study the process of process modeling, based on problem solving theories. Specifically, the work takes the approach that problems are first conceptualized as mental models, to which solution methods are applied. The paper suggests that investigating these two phases can help understand and hence improve semantic and syntactic quality of process models. The paper reports on an empirical study addressing the mental model created during process model development, demonstrating the feasibility of such studies. It then suggests designs for other studies that follow this direction.