Causal modeling is examined for its heuristic value by describing its use in an introductory course in educational evaluation, and presenting a student-developed model. Subjects included students in either master's programs in statistics or doctoral programs in administration and curriculum. Before preparing an evaluation plan and path analysis, students were cautioned that: (1) development of causal models with observational data requires a close coordination of subject matter theory and the statistical model, large samples, and "checks and tests" against other types of evidence; and (2) the model should be kept simple and only variables amenable to change or intervention should be used. A causal model specifying variables expected to influence end-of-year achievement test scores in mathematics for grades 4-6 was designed. In the model, the school had adapted a new approach to mathematics, i.e. the "manipulative" approach. Variables that are amenable to intervention were used, and included parental and students' attitudes toward mathematics, and the degree of implementation of the manipulative approach. The specification of the model focuses on what the school can do to influence educational outcomes, rather than assuming that status characteristics will account for end-of-year performance. (RL)