The effect of labels on nonlinguistic representations is the focus of substantial debate in the developmental literature. A recent empirical study (Twomey & Westermann, under review) suggested that labels are incorporated into object representations, such that infants respond differently to objects for which they know a label relative to unlabeled objects. However, these empirical data cannot differentiate between two recent theories of integrated label-object representations, one of which assumes labels are features of object representations, and one which assumes labels are represented separately, but become closely associated with learning. We address this issue using a neurocomputational (autoencoder) model to instantiate both theoretical approaches. Simulation data support an account in which labels are features of objects, with the same representational status as the objects’ visual and haptic characteristics.