Sound　rendering　is　a　technique　which　uses　auditory　display　to　communicate　information　to　a　user　and　provides　an　alternative　way　of　visualization.　Traditional　3D　sound　rendering　techniques　are　lower　effective　when　the　number　and　types　of　sound　sources　increase.　In　this　paper,　we　propose　a　method　of　3D　sound　source　management　based　on　osgAL　toolkit,　by　using　sound　classification　and　XML　document　structure　for　sound　properties.　We　allocate　buffers　for　necessary　sound　sources　and　cull　inaudible　sound　sources,　which　reduce　system＇s　consumption　of　time　and　space.　Experiments　on　train-driving　simulation　system　show　that　our　method　provides　a　realistic　and　effective　simulation.