Most studies of human category learning involve category structures that do not change, or that change in a way that is independent of people's categorization behavior. We consider the situation in which successful category learning causes categories to change. In an experiment, participants learned from feedback whether animals are healthy or diseased. Once their categorization accuracy was near-perfect, the category structure changed so that different animals became diseased. Based on exploratory data analysis and the application of two category learning models, we argue that, once they detect a category change, people retain what they have learned about healthy animals, but reset what they have learned about diseased animals. We discuss future modeling goals and emphasize the need for learning models to study situations in which people's behavior impacts the dynamics of the environment in which learning takes place.