We here evaluate the heterogeneous effects of newspaper endorsements of U.S. Presidential candidates in the 100 days preceding the 2008 and 2012 elections on the probability that they win the election. Our identification strategy relies on daily variations in the winning probabilities (obtained from the Intrade prediction market) and the fact that newspapers decide their endorsements weeks before their announcement. Endorsements that are classified as surprising and consistent have the largest effect. An endorsement is surprising when the newspaper has not traditionally endorsed the candidate's party. An endorsement is inconsistent when the newspaper leans ideologically to one party but endorses a candidate from another party. (JEL L82, D7)