This paper utilizes the feature of the CHDS data from New Zealand that children are sampled for extremely long individual histories of their class size experiences as well as their scholastic and early labor market outcomes. Our interest is to explore the full set of empirical implications of the recent experimental evidence on class size effects on student achievement from the United States in Project STAR for observational data. We argue that one implication of Project STAR is that only persistent class size reduction policies may have detectable effects, and so the longitudinal aspect of CHDS is necessary to detect class size effects. We account for the observational nature of the CHDS (in that children were not randomly assigned to different class sizes) by examining the long-run trends in test score growth, rather than levels. Consistent with the experimental evidence, we find statistically and economically significant effects of children being assigned to persistently smaller classes on both childhood test score growth as well as on early adult outcomes. Our analysis points the way towards the unification of experimental and observational evidence on class size effects, as well as highlighting several possible pitfalls in the analysis of observational data on this topic.