Within many theories of categorization, selective attention is described as a critical mechanism by which salient information is preferentially encoded during learning. Foundational model implementations of this idea estimate differences in encoding variability by specifying salience at the level of each dimension. We propose that overt attention during initial object presentation is a necessary condition for feature encoding, as well as subsequent retrieval of feature information at test. We present a data-constrained corollary to Medin and Shaffer's (1978) seminal context model that accounts for partial stimulus encoding, such that the storage of feature information is directly informed by gaze fixations to spatially-segregated dimensions. Using data from 232 preschool-age children to explore strategic influences of attention, we demonstrate that specifying which feature information is plausibly accessible within the category representation is sufficient for predicting recognition and categorization behavior.