In　this　paper,　as　physical　resources　of　sensor　nodes　are　limited,　such　as　storage　and　computing　power,　frequent　faults　and　reconfiguration　caused　by　failure　nodes　may　occur.　We　propose　a　multi-agent-based　adaptive　task　allocation　algorithm　in　wireless　sensor　network　(WSNs).　The　algorithm　applies　multi-agent　theory　and　technology　to　the　adaptive　task　allocation　in　WSNs　to　recover　the　network　with　the　minimum　cost.　Simulation　experiment　demonstrates　that　unexecuted　tasks　of　unavailable　nodes　can　effectively　migrate　to　other　health　nodes　in　WSNs　by　the　algorithm,　and　the　experiment　results　show　that　the　algorithm　can　save　more　energy　and　help　to　improve　the　performance　of　the　whole　network.　©2009　IEEE.