Research has shown that simulations that are more realistic provide better knowledge transfer and richer educational experiences. Consequently, the challenge for simulation designers has been to "create a more realistic model." Unfortunately, the concept of realism is a subject of much debate and differing understanding. At the lowest level, realism deals with the physical representation of entities within the training space, i.e. do they look right. Simulation designers have attacked this level of realism with success. However, higher order concerns of realism are often overlooked in simulation design. Entities may look real, but they must also act real, and act real for realistic reasons. Unfortunately, these first and second order effects are often hard to detect during requirements analysis. Simulation designers have not yet established a structured method for determining them. However, they are critical to produce a good simulation for use in a context based learning environment, especially for those used in training decision-making skills. The identification of realism concerns will have an impact on a number of design, control, and interoperability concerns as well. This paper proposes a structured approach for identifying and organizing training system requirements that address the physical, first, and second order concerns of realism.