Pré-publication, Document de travail: Individual-level discrimination is often attributed to deep-seated prejudice that is difficult to change. But at the societal level, we sometimes observe rapid reductions in discriminatory preferences, suggesting that social interactions and the communication they entail might drive such shifts. I examine whether discrimination can be reduced by two types of communication about a minority: (i) horizontal communication between majority-group members, or (ii) top-down communication from agents of authority (e.g., the legal system). I run a field experiment in urban India (N=3,397) that measures discrimination against a marginalized community of transgender people. Non-transgender participants are highly discriminatory: in a control condition, they sacrifice 1.9x their daily food expenditure to avoid hiring a transgender worker to deliver groceries to their home. But horizontal communication between participants sharply reduces discrimination: participants who were earlier involved in a group discussion with two of their neighbors no longer discriminate on average, even when making private post-discussion choices. This effect is 1.7x larger than the effect of top-down communication that informs participants about the legal rights of transgender people. The discussion’s effects are not driven by virtue signaling or correcting a misperceived norm. Instead, participants appear to persuade each other to be more pro-trans, partly because pro-trans participants are the most vocal in discussions.