According　to　the　communication　capability　of　current　sensor　nodes,　a　novel　scheme　(REDD)　is　proposed　for　data　delivery　in　large-scale　wireless　sensor　networks.　In　REDD,　the　clustering　network　architecture　is　formed　by　periodically　selecting　cluster　heads　among　sensor　nodes　through　dynamic　competition.　The　intra-cluster　communication　is　based　on　TDMA　scheduling,　in　which　a　member　node　transmits　its　sensory　data　through　an　independent　intra-cluster　channel　to　its　cluster　head　in　the　time　slot　assigned　by　the　cluster　head.　The　inter-cluster　communication　is　based　on　CSMA　competition,　in　which　a　connected　overlay　network　is　constructed　among　all　cluster　heads　to　forward　the　aggregated　data　through　multi-hops　to　the　sink.　Simulation　results　show　that,　compared　with　the　existing　work,　the　REDD　can　achieve　a　better　clustering　performance　with　lower　costs,　and　effectively　reduce　the　end-to-end　delay　for　reliable　data　transmission.