The Milwaukee voucher program, as implemented in 1990, allowed only nonsectarian private schools to participate in the program. However, following a Wisconsin Supreme Court ruling, the program was expanded to include religious private schools in 1998. This second phase of the voucher program led to more than a three-fold increase in the number of private schools and almost a four-fold increase in the number of choice students. Moreover, because of some changes in funding provisions, the revenue loss per student from vouchers increased in the second phase of the program. This paper analyzes, both theoretically and empirically, the effects of these changes on public school performance (as measured by test scores) in Milwaukee. It argues that voucher design matters and that the choice of parameters in a voucher program is crucial in determining the effects of public school incentives and performance. In the context of a theoretical model of public school and household behavior, the paper establishes that the policy changes will lead to an improvement of the public schools in the second phase of the program as compared with the first phase. Following Hoxby (2003a, 2003b) in the classification of treatment and control groups and using 1987-2002 data and a difference-in-differences estimation strategy in trends, the paper then shows that the theoretical prediction is validated empirically. This result is robust to alternative samples and specifications and survives robustness checks, including correcting for mean reversion. A technical appendix is included. (Contains 15 tables, 3 figures and 50 footnotes.)