In　this　paper,　a　hybrid　music　recommendation　system　is　proposed,　which　combines　collaborative　filtering　and　content-base　recommendation.　Neither　of　these　two　parts　can　make　full　use　of　all　the　information.　Our　method　integrates　both　user　rating　and　music　content　information　using　an　expansion　method　of　LSA　(Latent　Semantic　Analysis)　called　M-LSA.　We　use　a　text　representation　for　music　content　information,　which　is　obtained　by　K-means　Clustering　or　HMM　method.　Experiments　on　the　data　of　300　popular　songs　show　that　the　proposed　approach　achieves　satisfactory　results.