The extent to which people learning categories generalize should depend in part on their beliefs about how the instances were sampled. Bayesian models of sampling have been successful in predicting that generalization can decrease as more instances of a category are encountered. This has only been shown in tasks were instances are all from the same category, but contrasts with the predictions from most standard models of categorization that predict when multiple categories exist, people are more likely to generalize to categories that have more instances when distances between categories is controlled. In this current work we show that in both one- and two-category scenarios, people adjust their generalization behavior based on cover story and number of instances. These patterns of generalization at an individual level for both one- and two-category scenarios were well accounted for by a Bayesian model that relies on a mixture of sampling assumptions.