Although the framing of public opinion has often been conceptualized as a collective and social process, experimental studies of framing have typically examined only individual, psychological responses to alternative message frames. In this research we employ for the first time group conversations as the unit of analysis (following Gamson 1992) in an experimental study of framing effects. Two hundred and thirty-five American citizens in 50 groups (17 homogeneously conservative groups, 15 homogeneously liberal groups, and 18 heterogeneous groups) discussed whether or not gay and lesbian partnerships should be legally recognized. Groups were randomly assigned to one of two framing conditions (a "homosexual marriage/special rights" frame or a "civil union/equal rights" frame). Results indicated framing effects that were, in all cases, contingent on the ideological leanings of the group. The "marriage" frame tended to polarize group discussions along ideological lines. Both liberal and conservative groups appeared to find their opponents' frame more provocative, responding to them with a larger number of statements and expressing greater ambivalence than when reacting to more hospitable frames.