Recommender Systems (RSs) are tools focused on suggesting items that match the interests and preferences of a target user. They have been used in several domains such as e-commerce, e-learning, and social networks. These systems require the elicitation of user preferences, which are not always precise because there are external factors such as human errors, or the inherent vagueness associated to human beings; which are usually related to user behaviors. In RSs, such imperfect behaviors are identified as natural noise (NN), and can bias negatively the recommendation, which affects the RS performance. The current chapter presents two fuzzy models for NN management in a flexible way, which guarantees robust modeling of the uncertainty associated to the user profiles. These models are conceived for individual and group recommendation scenarios respectively, as a data preprocessing step before the recommendation generation. Two case studies are developed to show that the proposals lead to improvements in the accuracy of individual and group recommender systems.