Today there is a growing interest in the combination of knowledge management techniques with the emotions, induced by music, to build Music Recommendation Systems. Advertisers can exploit music recommendations to select the type of music that can produce appropriate emotional reactions to consumers. The goal of our research is to specify an efficient and accurate way for classifying listeners’ emotions and recommend music tracks. Metadata and the emotions, induced by music, are utilized to implement a prototype of a low cost personalized Music Recommendation System. Experiments are conducted on a set of 1000 tracks with three classes of music emotions. The results indicate the classification algorithm that can better predict the emotions evoked by a song, based on associated acoustic metadata. Eventually some conclusions are given.