Our primary aim is to forecast, rather than explain, presidential election results, using aggregate time series data from the post-World War II period. More particularly, we seek prediction of the presidential winner well before the election actually occurs. After comparing the performance of several naive blvariate models based on economic performance, international involvement, political experience, and presidential popularity, we go on to formulate a multivariate model. This economy-popularity regression model rather accurately forecasts the winner 6 months in advance of the election, by employing spring measures of presidential popularity and the growth rate in real GNP per capita. Furthermore, the model's performance, both ex post facto and prior to the election, compares favorably with the Gallup final preelection poll taken only a few days before the election.