One problem with describing election vote shares using ordinary least squares (OLS) is that it ignores the possible presence of spatial error correlation, whereby the errors are correlated in a systematic manner over space. This omission can bias OLS standard errors. We examine the 2004 presidential county vote outcome using OLS and a spatial error model (SEM) that accounts for spatial autocorrelation in the error structure. We find that spatial error correlation is present, that the SEM is superior to OLS for making inferences, and that several factors deemed important to the 2004 election outcome are not significant once the spatial error autocorrelation is taken into account.