COVIS is a neurobiologically motivated model of human category learning. It hypothesizes that two neurobiologically distinct systems mediate different kinds of category learning tasks. The explicit system is based on cortical areas (e.g., prefrontal cortex, anterior cingulate cortex) and their connections to subcortical areas and is thought to mediate learning in tasks requiring hypothesis testing, logical reasoning, and executive attention. The procedural system is based on a collection of subcortical areas (the basal ganglia) and is believed to mediate non-declarative, procedural learning. Both systems of COVIS posit a critical role of dopamine in learning. Further, in its current formulation, both systems of COVIS learn independently and receive independent feedback signals. In the present report, a computational implementation of COVIS was described and used to explore dopamine in learning and to test a new assumption that the systems do not learn independently by delivering a single source of feedback to both systems of COVIS. First, COVIS was used to simulate the results of nine datasets with human subjects who are either dopamine deprived (e.g., Parkinson's disease patients) or dopamine abundant (e.g., induced positive mood). Overall, COVIS was able to account for each data set through only dopamine-related parameter manipulation. Second, by exhaustively searching the model's parameter space, it was shown that the procedural system of COVIS could not learn with a single source of feedback unless it controlled the overall output of the model for most of the simulation. Further simulations indicate that assuming the explicit system trains up (i.e., bootstraps) the procedural system mitigates its failure to learn with one source of feedback. It is hypothesized that this bootstrapping is made possible through recurrent corticostriatal projections from motor areas.