In the development of distance learning, advances in cognitive science merge with new technology to deliver instruction worldwide. However, one major difficulty in evaluating the efficacy of these tools is determining which elements of instruction truly lead to observed changes in student performance. As content, pedagogical methods, and media are intertwined, identifying the "active ingredients" is an essential element of facilitating training that is of high quality and minimizes development costs (Clark & Estes, 2000). To accurately evaluate applications in the field, researchers must be able to identify specific instructional components, make decisions on who and what will be subject to treatment, and accurately draw inferences regarding causal interactions without the control offered by a laboratory. The purpose of this paper is to briefly review current use of various research methods for evaluating instructional technologies, discuss previous solutions to balancing the conflicting demands of internal and external validity, and then to propose a new research design that achieves this goal in a manner compatible with many instructional technology applications.