Causal modeling was applied to data collected in a naturalistic setting in an attempt to validate a conceptual model of teaching. These data included supervisory ratings of the variables: specifying performance objectives, diagnosing learners, selecting instructional strategies, and interacting with learners collected on 82 student teachers. Additionally, the mean number of objectives achieved by the classes of each of the student teachers was used as a measure of the fifth variable in the model, evaluating the effectiveness of instruction. A recursive causal model which described the relationship of these variables was developed and analyzed, using four linear equations. Examination of the path coefficients from these equations revealed that the variables, specification of performance objectives and diagnosing learners account for three tenths of the variance in selecting instructional strategies. Other results indicated there were low to moderate amounts of variance shared by the variables of the model. This study illustrates the application of causal modeling techniques in testing theoretical models in education using data collected in naturalistic, non-experimental settings. (Author)