This study investigated the assumptions of prototype and exemplar models of human category learning, with a particular focus on the impact of category frequency. We used baseline and recency-weighted variants of prototype and exemplar models to examine the computational mechanisms underlying categorization decisions when one category was presented more frequently than the other. We employed extensive sets of stimuli derived from bivariate normal distributions and manipulated category frequency during training across four experiments using different category structures. In the transfer phases, participants classified novel stimuli. Across all studies, the results revealed a significant frequency effect, with participants showing a preference for categorizing novel items as members of the more frequently encountered category. This preference extended to transfer stimuli outside the trained region of the stimulus space. Model-based analyses indicated that the recency-weighted generalized context model exemplar model, which computes summed similarity via a Decay reinforcement learning rule, consistently outperformed other models in fitting the data and accurately reproducing the observed classification patterns across all experiments. Both prototype models failed to account for the observed frequency effects. While the baseline generalized context model was able to account for frequency effects, it did not capture recency effects. These findings suggest that relative category frequency influences human behavior when categorizing novel items. The computational modeling results revealed that evidence for categorization decisions is recency-weighted and accumulative rather than averaged. (PsycInfo Database Record (c) 2025 APA, all rights reserved).