Constitutive　relationship　of　as-cast　904L　austenitic　stainless　steel　is　comparatively　investigated　by　the　Arrhenius-type　constitutive　model　incorporating　the　strain　effect　and　back-propagation　(BP)　neural　network.　The　experimental　true　stress-true　strain　data　were　obtained　from　hot　compression　tests　on　the　Gleeble-1500D　thermo-mechanical　simulator　in　the　temperature　range　of　1000-1150　degrees　C　and　strain　rate　range　of　0.01-10　s(-1).　The　corrected　data　with　the　friction　and　the　temperature　compensations　were　employed　to　develop　the　Arrhenius-type　model　and　BP　neural　network　respectively.　The　accuracy　and　reliability　of　the　models　were　quantified　by　employing　statistical　parameters　such　as　the　correlation　coefficient　and　absolute　average　error.　The　results　show　that　the　proposed　models　have　excellent　predictabilities　of　flow　stresses　for　the　present　steel　in　the　specified　deformation　conditions.　Compared　with　the　Arrhenius-type　model,　the　optimized　BP　neural　network　model　has　more　accuracy　and　capability　in　describing　the　compressive　deformation　behavior　at　elevated　temperature　for　as-cast　904L　austenitic　stainless　steel.　(C)　2012　Elsevier　B.V.　All　rights　reserved.