Evaluation of the efficacy of school policies requires measures of student performance across schools and states, but conventional approaches to constructing the relevant data can be very misleading. this paper develops an approach to estimating marginal school effects at the state level. It then documents and estimates the magnitude of biases introduced by commonly employed estimators of school quality. Direst estimates of achievement growth, or value-added, are shown to be far superior to any alternative correction that is commonly employed. Especially at the state level, nonrepresentative data such as aggregate SAT scores provide very biased measures of school quality differences -- even when statistical adjustments for demographic differences and varying participation rates are employed.