Researchers in spatial cognition have debated for decades the specificity of the mechanisms through which spatial information is processed and stored. Interestingly, although rodents are the preferred animal model for studying spatial navigation, the behavioral methods traditionally used to assess spatial memory do not effectively test the predictions of specificity in their representation. To address such issues, the present study tested the ability of mice to use boundary geometry and features to remember a goal location across 2 types of tasks-a working memory task with a changing goal location, and a reference memory task with 1 rewarded goal location. We show for the first time that mice, like other animals, can successfully encode boundary geometry in a working memory spatial mapping task, just as they do in a reference memory task. Their use of a nongeometric featural cue (striped pattern), in contrast, was more limited in the working memory task, although it quickly improved in the reference memory task. We discuss the implications of these findings for future research on the neural and genetic underpinnings of spatial representations.