This article assesses the relative power of proximity and directional models in explaining party evaluation in 1993 and 1997 Canadian Election Study data. We first nest the competing models in an encompassing framework and then test the models under alternative representations of party location, where these alternatives lie at the heart of the debate. Fixing party locations at sample means favors directional outcomes, while allowing them to vary over individuals favors proximity outcomes. We treat these as bounds for the model and find that even when the setup favors directional results, the proximity model dominates. This leaves us with a puzzle in that Canadian parties adopt locations that cannot be accounted for by either model.