Computation　offloading　is　a　promising　way　to　improve　the　performance　as　well　as　reducing　the　battery　power　consumption　of　a　mobile　application　by　executing　some　parts　of　the　application　on　a　remote　server.　Recent　researches　on　mobile　cloud　computing　mainly　focus　on　the　code　partitioning　and　offloading　techniques,　assuming　that　mobile　codes　are　offloaded　to　a　prepared　server.　However,　the　context　of　a　mobile　device,　such　as　locations　and　network　conditions,　changes　continuously　as　it　moves　throughout　the　day,　and　there　are　multiple　options　of　cloud　resources,　including　remote　cloud　computing　services　and　nearby　cloudlets.　In　order　to　offload　computation　to　the　cloud　resource　with　powerful　processors　as　well　as　fast　network　connection,　it　needs　to　dynamically　select　the　appropriate　cloud　resource　and　then　offload　mobile　codes　to　it　at　runtime,　according　to　the　context　of　the　mobile　device　and　possible　cloud　resources.　In　this　paper,　we　present　a　framework　for　context-aware　computation　offloading.　First,　a　design　pattern　is　proposed　to　enable　an　application　to　be　computation　offloaded　on-demand.　Second,　an　estimation　model　is　presented　to　automatically　select　the　cloud　resource　for　computation　offloading.　Runtime　data　about　computation　tasks,　contexts　of　the　mobile　device　and　possible　cloud　resources　is　collected　and　modeled　at　client　side,　in　order　to　make　an　optimal　offloading　decision.　A　thorough　evaluation　on　two　real-world　applications　is　proposed,　and　the　results　show　that　our　approach　can　help　reduce　execution　time　by　6%-96%　and　power　consumption　by　60%-96%　for　computation-intensive　applications.　©　2016　IEEE.