Building intelligent environment is one of crucial challenges for ubiquitous computing developers. To make the environment adapt rationally according to the desire of users, the system should be able to guess users’ interest, by learning users’ behavior, habit or preference. While learning the user preference, dealing with uncertainty and conflict resolution is of the utmost importance. When many users are involved in a ubiquitous environment, the decisions of one user can be affected by the desires of others. This makes learning and prediction of user preference difficult. To address the issue, we propose an approach of user preference learning which can be used widely in context-aware systems. We use Bayesian RN-Metanetwork, a multilevel Bayesian network to model user preference and priority.