Scholars, policy makers, and the general public have expressed growing concern about the possibility of large-scale political violence in the United States. These worries find support in studies revealing that many American partisans support the use of violence against rival partisans. Here we propose that support for partisan violence is based in part on greatly exaggerated perceptions of rival partisans’ support for violence. We also predict that correcting these inaccurate “metaperceptions” can reduce partisans’ own support for partisan violence. We test these hypotheses in a series of pre-registered, nationally representative, correlational, longitudinal, and experimental studies (total n = 4,741), collected both before and after the 2020 U.S. Presidential election and the 2021 U.S. Capitol attack. In Studies 1 and 2 we found that both Democrats’ and Republicans’ perceptions of their rival partisans’ support for violence and willingness to engage in violence were very inaccurate, with estimates ranging from 239% to 489% higher than actual levels. Further, we find that a brief, informational correction of these misperceptions reduced support for violence by 37% (Study 3) and willingness to engage in violence by 44% (Study 4). In the latter study, a follow-up survey revealed the correction continued to significantly reduce support for violence approximately one month following the study. Together, these results suggest that support for partisan violence in the United States stems in part from systematic overestimations of rival partisans’ support for violence, and that correcting these misperceptions can durably reduce support for partisan violence in the mass public.