Research has shown that students’ epistemic beliefs influence the way they learn, think and reason in any given context (Schommer-Aikins, 2004). However, in the science learning context, the relationship between the level of epistemic sophistication, learning, and learning outcomes is sometimes ambiguous (Elby & Hammer, 2001). Taking this result as a point of departure, we examined the relationships between students’ scientific epistemic beliefs (SEB), their approaches to a computer simulated task, and the quality of their solutions. 19 tenth grade students, with different SEB, were selected to participate in a constructionist computer-simulation in classical mechanics. Constructionist learning environments emphasize the scope for students’ to take control of their own learning, draw their own conclusions, and use their own knowledge in order to construct objects (Harel & Papert, 1991). Students’ manipulations of the simulation and any spoken comments were video-recorded and subsequently coded by an inductive approach. Relationships between students’ SEB and problem solving quality were explored by hierarchical orthogonal partial least squares analysis. The results revealed that different sets of SEB were conducive to different aspects of students’ problem solving process and outcomes. Theoretically sophisticated beliefs were in general associated with logical strategies and high solution complexity. However, our results suggest that there might not be a universal relationship between the degree of theoretical sophistication of students’ SEB and quality of learning outcomes. The relationship can only be understood in terms of the actions they induce, and the results of these actions. It is therefore of great importance to further explore the productiveness of SEB in different types of learning situations.