We examine empirical implications of Lazear's (2001) educational production model. Using exogenous variation on course scheduling for ninth graders in Chicago Public Schools (CPS), we study heterogeneity in the impact of class size on student achievement in reading and mathematics. Our identification strategy allows us to analyze an underlying mechanism by which class size affects student performance, the behavioral composition of a classroom. Our classroom composition measure is constructed as the average number of non-disruptive students in attendance on a given school day; we characterize a classroom as being comprised of disruptive and non-disruptive students, where a student is considered disruptive if he or she dropped out of high school in any grade. Our data are taken from CPS administrative student transcript files, which include the ordering of classes over the day, student absences, course titles, grades, scores from standardized tests in reading and mathematics, and demographic characteristics for the universe of CPS high school students from the 1993-94 to the 2005-06 school years. To study the effects of the behavioral composition of a classroom on academic achievement, we use an instrumental variables approach, exploiting exogenous variation in the period of the day a course is offered in CPS. Our analysis focuses on students' course passing and test scores in four ninth grade course subjects: regular English I, remedial English I, regular Algebra I, and remedial Algebra I. Consistent with the Lazear framework, we find that an additional non-disruptive student in attendance increases the probability of passing English I and Algebra I, with larger effects for students in remedial versus regular classes. For regular English I students, we estimate a positive relationship between the number of non-disruptive students in attendance and own reading test score.