Two experiments investigated the production and comprehension of referring expressions that contain a negative property ("the marker without a cap"). Experiment 1 showed that participants do use negative properties in their object descriptions, but that they were almost always accompanied by other properties, leading to referential overspecification. In experiment 2, participants identified objects based on descriptions that contained negative properties. While participants were faster in identifying objects that were described with preferred properties such as color, response times for objects described with a negative property ("the marker without a cap") and a positive property ("the marker with a cap") did not differ. The results provide behavioral grounds for extending referring expression generation algorithms to include negative properties.