RNA　molecules　play　important　roles　in　biological　processes,　their　functions　are　intimately　related　to　structural　dynamics.　Elastic　network　model　(ENM)　has　achieved　great　success　in　predicting　the　large-amplitude　collective　behavior　of　proteins.　However,　for　loosely-packed　RNA　structures,　ENM　models　can　not　reproduce　their　dynamics　as　accurate　as　the　densely-packed　ones.　In　this　work,　the　multiscale　Gaussian　network　model　(mGNM)　is　extended　to　predict　dynamic　properties　of　RNAs.　All　tests　are　performed　on　a　non-redundant　RNA　structure　database　we　constructed.　In　results,　for　B-factor　reproduction,　encouragingly　mGNM　achieves　a　significant　improvement　with　the　average　value　of　Pearson　correlation　coefficient　(PCC)　between　theoretical　and　experimental　B-factors　being　0.732,　much　higher　than　0.494　and　0.321　obtained　by　conventional　GNM　and　parameter-free　GNM　(pfGNM)　models,　respectively.　Furtherly,　mGNM　attains　a　larger　improvement　in　B-factor　prediction　for　loosely-packed　parts.　Additionally,　based　on　the　analysis　of　functional　movements,　mGNM　can　properly　make　domain　decompositions　for　tRNA(Asp)　and　xrRNA.　This　work　can　strengthen　the　understanding　of　the　intrinsic　dynamics　of　RNAs,　and　mGNM　is　expected　to　have　a　bright　prospect　in　dynamic　analyses　for　loosely　folded　biomolecules,　especially　RNAs.