There are many types of programs for Spanish speaking students in the US, with varying methods and goals. Some preliminary work suggests that bilingual classrooms may differ widely in instruction, even under the same program labels. However, there are few studies which have compared the extent to which various bilingual program models differ in actual instruction delivered. Directly measuring instructional practice however, is difficult and costly, involving the influence of time, raters, content, and programs (Raudenbush, 2008). The purpose of the current paper is to estimate the relative influence of these important sources of variance in classroom observations completed in a large quasi-experiment of bilingual education. (Contains 6 tables.)