Signal detection theory forms the basis of many current models of memory, choice, and categorization. However, little research has examined precisely how the decisionmaking process unfolds over time. In this paper, a new nonparametric, dynamic model is proposed with the intentions of ameliorating some long-standing issues in the signal detection framework and describing the changes in signal detection performance over time. The model uses a recursive kernel density estimation procedure that accumulates and stores experience across trials. I present the results of several simulations and show that the proposed model bypasses the rigid assumptions of prior internal representations of the sampling distributions and as a consequence, it allows the criterion location to shift to accommodate new information as it is presented.