Interventions aimed at changing attitudes are often focused on individual responses. However, attitudes do not shift in isolation; individuals are embedded in rich social networks that can reinforce, push against, or emulate changes. We conducted a field experiment in 16 villages in northwestern Uganda which randomly assigns a perspective-taking treatment aimed at reducing prejudice towards refugees to 40% of the households in each village. Our design includes a measure of full household social networks as well as measures of individual attitudes at baseline, immediately after treatment, and at endline after a two week interim in which people were free to discuss the issue with others in their village. We find that the treatment does warm attitudes of the treated on average in the short-term, though with considerable variance. We also find that people’s attitudes change in the longer-term based on informal conversations with others in the network after treatment. By the endline, the control attitudes warm on average too, consistent with classical spillovers. Inconsistent with classical spillovers, the treated attitudes warm even further, and the ultimate attitudes of the control are a function of not just the presence of treated network neighbors but these neighbors’ individual reactions to the treatment. We argue that the results are consistent with a period of “social processing” in which revealed reactions ultimately shape the attitudes of both the treated and the control. We stipulate a simple model of such a process and show that it can generate non-classical spillovers like those we observe. Taken together, these findings show the importance of understanding the social process that can reinforce or unravel individual-level attitude change; it appears essential to designing interventions with a lasting effect on attitudes.