How a person interprets music and what prompts a person to feel certain emotions are two very subjective things. This dissertation presents a method where a system can learn and track a user's listening habits with the purpose of recommending songs that fits the user's specific way of interpreting music and emotions. First a literature review is presented which shows an overview of the current state of recommender systems, as well as describing classifiers; then the process of collecting user data isdiscussed; then the process of training and testing personalized classifiers is described; finally a system combining the personalized classifiers with clustered data into ahierarchy of recommender systems is presented.