Physics engines have been used in robotics research for a long time. Beside their traditional application as a substitute for real world interactions due to their higher speed, safety and flexibility, they have recently also been used for motion planning and high level action planning. We propose to further explore the idea of using a physics engine as means to give a robot a basic physical understanding of its environment. In this paper, as a preliminary step, we study, how accurately the process of pushing flat objects across a table with a robot arm can be predicted in a physics engine. We also present an approach to adapt the engines parameters to enhance the simulation accuracy.