This paper examines the construction and use of visual representations for reasoning. Visual representations have often been categorized by a resemblance criterion: representations of objects are considered iconic, because they resemble a perceived object, while representations of abstractions, which cannot resemble an object, are considered unconstrained and therefore arbitrary. In contrast, I argue that visual representations used in reasoning exhibit a logic that supersedes resemblance. Humans using visual representations in reasoning map concepts onto spatial structures metaphorically according to a simple but structurally and directionally sensitive rule: features map to features and relations map to relations. This rule can be observed in reasoning tasks with children and adults, and is offered as an explicit heuristic to guide the construction and use of visuospatial representations in reasoning.