This paper examines the impact of vouchers in general and voucher design in particular on public school performance. It argues that all voucher programs are not created equal. There are often fundamental differences in voucher designs that affect public school incentives differently and induce different responses from them. It analyzes two voucher programs in the United States. The 1990 Milwaukee experiment can be looked upon as a "voucher shock" program that suddenly made low-income students eligible for vouchers. The 1999 Florida program can be looked upon as a "threat of voucher" program, in which schools getting an "F" grade for the first time are exposed to the threat of vouchers, but do not face vouchers unless and until they get a second "F" within the next three years. In the context of a formal theoretical model, the study argues that the threatened public schools will unambiguously improve under the Florida-type program, and this improvement will exceed that under the Milwaukee-type program. Using school-level scores from Florida and Wisconsin and a difference-in-differences estimation strategy in trends, it then shows that these predictions are validated empirically. These findings are reasonably robust in that they survive sensitivity checks including correcting for mean reversion and a regression discontinuity analysis. (Contains 12 tables, 3 figures and 62 footnotes.)