Local　cosine/sine　basis　is　a　localized　version　of　cosine/sine　basis　with　a　smooth　window　function.　It　has　orthogonality　and　good　time　and　frequency　localization　properties　together　with　a　fast　algorithm.　In　this　paper,　we　present　a　new　method　combining　the　main　idea　of　local　cosine/sine　bases,　multi-scale　decomposition　and　the　dispersion　relation　to　form　a　multi-scale,　multi-dimensional　self-similar　dreamlet　transform.　Meanwhile,　a　storage　scheme　based　on　the　dreamlet　decomposition　is　proposed　by　using　zig-zag　sequence.　We　apply　this　method　to　the　SEG-EAGE　salt　model　synthetic　poststack　data　set　for　data　compression.　From　the　result,　almost　all　the　important　features　of　the　data　set　can　be　well　kept,　even　in　high　compression　ratio.　Using　the　reconstructed　data　for　migration,　we　can　still　obtain　a　high　quality　image　of　the　structure.　Through　comparing　with　other　decomposition　and　compression　schemes,　we　believe　that　our　sheme　is　more　closely　related　to　the　physics　of　wavefield　and　has　a　better　performance　for　seismic　data　compression　and　migration.　©　1996-2018　Society　of　Exploration　Geophysicists　All　Rights　Reserved.