Are labels cues to category membership or simply highly salient features? This question is difficult to answer definitively because of the challenge in identifying empirical predictions that would be distinct in each case: either way, one would expect labels to be highly interesting, easy to process, and preferentially used as the basis of generalization. Here we suggest that one difference should be in how the label directs (or fails to direct) attention to the other, less-salient features of the object. We perform a categorization experiment with complex objects containing many low-salience features, and find that labels affect attention to the other features in the same way that highly salient features such as color or sounds do (and unlike an explicit cue to category membership). This results in a diminished ability to use the less-salient features of the categories to generalize appropriately.