In many democracies, parties use primary elections to nominate candidates. Primaries may help parties select quality candidates, but they can also expose flaws and offend losing candidates’ supporters. Do divisive primaries help or harm parties in the general election? Existing research is mixed, likely because of issues of selection and omitted variables. We address these issues by using U.S. states with runoff primaries—second-round elections which, when triggered, create more divisive primaries. Using a regression discontinuity design, we estimate that going to a runoff decreases the party’s general-election vote share in the House and Senate by 6–9 percentage points and decreases the party’s win probability by 21 percentage points, on average. Opposing results in state legislatures suggest that divisive primaries are damaging when salience is high but beneficial when it is low, a pattern we argue is driven by the competing effects of information in high vs. low salience primaries.