Maintaining that the generalizability of behavioral measures has not been sufficiently established to permit conclusions about the relationship between teacher behavior and student achievement, the present research examines the generalizability of classroom interaction variables measured by the Brophy-Good Teacher-Child Dyadic Interaction System during 2nd and 3rd grade reading instruction. Using generalizability theory as the statistical basis for data analysis, the number of measurement occasions required to reach the 0.7 level of generalizability for five clusters of classroom interaction variables were identified. Analyses revealed that the interaction characterizing reading instruction differs from that characterizing other kinds of instruction in regard to: 1) proportion of public to private teacher-pupil interactions; 2) nature of questions asked; and 3) teacher behavior concerning feedback, pupil involvement, and question difficulty. (Author)