Many cite the growing tendency of political parties to nominate ideologically extreme candidates in U.S. primary elections as a reason for increasing congressional polarization. However, a lack of quantitative data on candidate ideology makes this claim difficult to test. We propose a unique solution to this problem that exploits data from the increasingly popular realm of social media by estimating ideal points for candidates for the U.S. Senate based on the patterns of connections in their Twitter social network. We identify a latent ideological dimension from the matrix of following relations, which corresponds closely to roll call–based estimates for congressional officeholders. Controlling for other relevant factors, we find support for the hypothesis that ideological extremity is advantageous in party primaries.