Much presidential election forecasting research employs macromodels based on national economic and political fluctuations. Micromodels based on surveys of individuals exist, but they are almost entirely pre-election explorations of vote intention. What has been neglected are micromodels derived from vote expectations. We show, by analysis of the American National Election Surveys 1956–1996, that voters themselves can forecast who will win the presidential election. We go on to explain some of sources of this forecasting ability, and to evaluate its precision. Voter forecasting models emerge as a useful alterative to current approaches.