Young children expect members of a familiar category to share nonobvious properties. The present experiments were designed to determine whether children expect unfamiliar objects with the same novel label to share nonobvious properties. In three experiments, 4-and 5-year-old children were taught properties about unfamiliar objects (e.g., a gnu-like animal) and then were tested on their generalization of each property to four test objects. The test objects were either similar or dissimilar in appearance to the target and received either the same or a different novel label from that of the target. When category labels and appearances were completely independent of one another, children drew inferences based on appearances rather than novel category labels. However, when the same pictures received familiar labels (e.g., a cow, a deer) instead of novel labels (e.g., a fep, a wug), children drew inferences based on category labels as well as appearances. Furthermore, children did draw inferences on the basis of novel labels when perceptual cues were not completely discrepant from the labels. These results suggest that novel labels help guide children's inferences from unfamiliar categories, but only when the conceptual basis of the names is clear. © 1990.