Recall of remembered locations reliably reflects a compromise between a target's true position and its region's prototypical position. The effect is quite robust, and a standard interpretation for these data is that the metric and categorical codings blend in a Bayesian combinatory fashion. However, there has been no direct experimental evidence that the 2 codings are actually combined. That is, at least 2 mechanisms can produce biased mean responses: (a) people may in fact take a weighted average of the metric and categorical representations, but (b) these 2 codings may instead compete for response, each winning with a certain probability. The present work investigated these 2 hypotheses for the cause of category-based distortions using a new distribution analysis. Participants viewed a target within a blank circle and reproduced its location after a short delay. The error data for individual participants were fit with a kernel curve, which provides a distribution without the assumption of normality. Almost all individual distributions displayed a clear biased main peak, indicating a weighted average between the representations, not an alteration between the 2 representations.