Although a large body of research has been produced both on semi-presidential regimes and patterns of electoral change from general elections to midterm or other non-general elections, the study of presidential elections in semi-presidential regimes remains, to quote one of the few exceptions in this regard, “uncharted territory” in the political science literature. Using a dataset on election results and turnout levels in all semi-presidential democracies since 1945, we test several hypotheses about changes in turnout levels and government parties and coalitions’ gains and losses in presidential elections. We show that while semi-presidential democracies with weaker presidencies do approximate the patterns predicted by the “second-order” model, that is clearly not the case where presidents hold more considerable powers, where government losses are explained by “negative voting” and “balancing” theories. The implications of these findings for the very definition of “semi-presidentialism” and the consequences of these regimes are also discussed.