The generalizability issue is reviewed through previously published results and by presenting new data. When classes (teacher-course combinations) are used as the object of measurement, studies show very consistent results. However, when either the teacher or the course is treated as the object of measurement, with the other being a facet over which generalization is desired, results are highly inconsistent, especially with regard to the effects of discipline and item content. The hypothesis that discrepancies are largely due to sampling fluctuations in variance component estimates is presented. For teachers and courses as objects of measurement, a closer look at item content and discipline is needed; also necessary, however, is an investigation of course, teacher, and student characteristics, and teaching/learning styles. The pooling of institutional data is one possible way of moving closer to definitiveness. (Author/GK)