Prior research suggests that the benefits from smaller classes may vary along multiple dimensions. In this paper we develop a flexible model of education production that incorporates the classroom-level time lost to disruption and the rate of learning during productive time as a function of teacher quality and individual propensity to acquire knowledge. We then investigate heterogeneity in class size effects by school poverty share, family income, teacher experience, and achievement percentile using data from Project STAR. We find that the benefits of small classes are consistently higher in schools with a larger low-income enrollment share. Conditional on school poverty share, we find little or no evidence that lower-income or lower-achieving students tend to realize larger benefits of smaller classes. Instead, we find that the return to smaller classes tends to increase with achievement regardless of school poverty share. Given the generally higher levels of disruption reported in higher poverty schools, this set of findings is consistent with, though not direct evidence of, the notion that reduced time lost to disruption is a primary mechanism through which smaller classes raise achievement and a compelling explanation for the empirical finding that class-size effects tend to be larger for lower-income children.