Original　innovations　are　important　for　the　technical　progress　and　the　economic　growth　of　countries　and　constructing　models　to　reveal　their　inherent　factors,　the　interaction　between　those　factors　and　their　specific　effects　is　becoming　quite　necessary.　This　paper　proposed　a　multi-agent　model　on　original　innovations　from　the　knowledge　views.　It　analysed　the　relationships　among　the　supporting　knowledge　system　of　the　original　innovation,　its　participators　and　the　scientific　collaboration　network,　designed　agents＇　behaviours　as　well　as　their　adaption　rules　and　simulated　the　model　on　Swarm　platform.　It　showed　that　the　system　with　very　few　research　directions　tends　to　have　the　best　innovation　performance.　As　for　systems　with　middle　or　higher　amount　of　research　directions,　adopting　the　strategy　prior　to　adapt　original　innovations　initiated　by　other　researchers　as　well　as　the　strategy　to　introduce　collaborations　between　experts　may　both　improve　the　system　innovation　performance.　These　two　strategies　all　have　their　individual　advantages　and　it　cannot　be　stated　which　one　is　always　better　than　the　other.　This　multi-agent　model　on　original　innovations　is　operable　and　applicable　and　the　deductions　drawn　from　its　simulation　results　can　be　supported　by　practice　and　may　contribute　to　our　cognition　of　original　innovations　and　their　improvement　strategies.　©　2013　Asian　Network　for　Scientific　Information.