The goal of this paper is to evaluate different approaches for deriving respondents' electoral utilities. The conventional approach draws on subjective distances between parties and respondents in low dimensional policy space, such as the left-right dimension. We compare this approach to more objectively revealed utilities as measured by scaling respondents' preferences on a wide range of policy issues, and the placement of parties by iterative expert surveys. In our analysis, we compare the classification performance of these approaches in the context of directional and proximity models of issue voting using statistical learning techniques. We argue that objectively revealed electoral utilities present a more nuanced picture that is as good as, if not superior, to traditional utilities, which tend to suffer from endogeneity.