In recent years, the rhetoric surrounding criminal justice policy has increasingly emphasized reform, rather than being "tough on crime." Although this change in rhetoric is aimed at building public support for reform, little is known about its efficacy. To test the efficacy of reform rhetoric, I conducted an Internet experiment using Amazon Mechanical Turk. Respondents were randomly assigned to one of six message conditions or to a control condition (no message) and then asked their views about eliminating the use of incarceration for select nonviolent offenses. Results from ordinal logistic regression models suggest that message frames that appeal to a respondent's self-interest or emphasize the unfairness of the punishment (not who is punished) tend to be most effective.