We leverage a unique two-stage experiment that randomized access to private school vouchers across markets as well as students to estimate the revealed preference value of school choice. To do this, we estimate several choice models on data only from control markets before turning to the treatment data for model validation. This exercise reveals that a model where school choice is constrained by ability-to-pay achieves better out-of-sample fit but still underpredicts experimental take-up of the voucher offer. We then present evidence from treatment markets that: a) the voucher offer also induced search; and b) private schools used program surplus to incentivize enrollment. Further, we show that a unified model incorporating these features can explain both the control and treatment data patterns. Estimates from that model imply that a targeted voucher program would have a marginal value of public funds (MVPF) of at least 3.