Although deliberation has a central place in democratic theory, scholars know little about how it actually works. Most deliberative theorists emphasize the many good consequences of deliberation. By contrast, Mansbridge suggests that deliberation in certain circumstances may exacerbate conflict. Scholarship on racial politics suggests that each hypothesis is complicated by implicitly racial language. Using a quasi-experiment, we contrast the rhetoric in two town meetings about school desegregation: a segregated meeting with homogeneous interests, in which segregated Whites unanimously argued against desegregation, and an integrated meeting with heterogeneous interests, in which segregated Whites argued against integrated Whites, Hispanics, and African Americans. We find that (a) deliberation at the segregated meeting maintained consensus among segregated Whites; (b) these citizens used coded rhetoric that appeared universal, well-reasoned, and focused on the common good, but in fact advanced their group interest; (c) deliberation at the integrated meeting maintained the conflict between segregated Whites and others; and (d) there, rhetoric that seemed universal to segregated Whites was decoded by the integrated audience as racist and group interested. We highlight the problem posed by the contested meaning of language and suggest ways to make deliberation more effective.