Spatial models of electoral competition locate voters and parties at points in euclidean space—representing issue positions—and specify utility of voters for parties as functions of these positions. Utility functions may also have stochastic components unassociated with issues. In this article probabilistic models are compared in which the utility function incorporates distance between voter and party positions (proximity model) or a scalar product (directional model). Model specification is significant because of its relation to party strategy and the resulting spatial distribution of parties. Maximum likelihood is used to estimate parameters of a mixed directional and proximity model—with stochastic and strategic components—from data in Norwegian and Swedish election studies. Expected spatial distributions of voters by party support are determined for the multiparty electorates of Norway and Sweden. Unlike previous deterministic work, which strongly favors the directional model, the results obtained here suggest that a mixture of proximity and directional probabilistic models may provide a substantially better fit than either pure model or a deterministic model.