Natural categories seem to be comprised of clustered stimuli that contain a myriad of correlated features; birds, for example, tend to fly, have wings, lay eggs, and make nests. Nonetheless, the evidence that people use these correlations during intentional category learning is overwhelmingly negative (Murphy, 2002). People do, however, show evidence of correlational sensitivity during other types of category learning tasks (e.g., feature prediction). The usual explanation is that intentional category learning tasks promote rule use, which discards the correlated feature information; whereas, other types of category learning tasks promote exemplar storage, which preserves correlated feature information. However, all of the intentional category learning tasks employed to examine correlational sensitivity to date have only used deterministic mappings of stimuli to categories (i.e., each stimulus belongs to only one category). The current thesis is concerned primarily with the effects introducing the probabilistic assignment of stimuli to categories on the acquisition of different types of correlational knowledge. If correlational knowledge depends on whether or not people selectively attend to the correlation then probabilistic reinforcement, which is predicted to increase attention shifting (Kruschke & Johansen, 1999), should lead to increased correlational sensitivity. The first paper of this thesis confirms that selective attention provides a way to explain the presence or absence of correlational knowledge in different tasks. However, selective attention models have been unable to account for tasks in which people use the correlation between a non-relevant cue and regions of the category space to switch between the application of multiple rules. This phenomenon, known as knowledge partitioning, is explored in the second paper of this thesis. This thesis also extends the empirical implications of the first two papers to existing research (see included paper 3) and also provides recommendations of how utilize this conceptualization of knowledge for practitioners in the applied setting (see included paper 4). Finally, in addition to increasing attention shifting, probabilistic feedback is also assumed to result in an attenuation of learning over time (Kruschke & Johansen, 1999); the fifth paper in this thesis provides empirical confirmation that people attenuate learning in response to unavoidable error.