An　adaptive　refinement　algorithm　for　motion　vector　in　video　transcoding　is　presented.　Making　use　of　the　spatial　coherence　and　amplitude　intensity　of　local　motion　vectors,　a　motion　activity　model　is　constructed,　by　using　which　the　search　range　of　the　macroblock　is　adaptively　confirmed.　Then,　a　multi-pattern　refinement　strategy　based　on　the　horizontal　and　vertical　search　(HAVS)　and　diamond　search　(DS)　is　proposed.　Simulation　results　show　that　the　proposed　method　reduces　the　computational　complexity　significantly　without　degrading　the　video　quality.　©　2010　IEEE.