Recommender systems have played a relevant role in e-commerce for supporting online users to obtain suggestions about products that best fit their preferences and needs in overloaded search spaces. In such a context, several authors have proposed methods focused on removing the users’ inconsistencies when they rate items, so-called natural noise, improving in this way the recommendation performance. The current paper explores the use of rating regularities for managing the natural noise in collaborative filtering recommendation, having as key feature the use of fuzzy techniques for coping with the uncertainty associated to such scenarios. Specifically, such regularities are used for representing common rating patterns and thus detect noisy ratings when they tend to contradict such patterns. An experimental study is developed for showing the performance of the proposal, as well as analyzing its behavior in contrast to previous natural noise management procedures.