A　novel　internal　model　control　strategy　is　presented　based　on　the　improved　extreme　learning　machine.　Specifically,　the　improved　extreme　learning　machine　is　used　to　estimate　the　internal　model,　and　internal　model　is　expanded　withfirst-order　term　by　Taylor　series　to　calculate　the　controller　for　the　control　system　indirectly,　which　avoids　calculating　the　inverse　for　internal　model　directly.　Moreover,　the　system　stability　and　error　are　also　analyzed　for　this　control　system　with　model　errors　and　disturbance.　The　proposed　control　strategy　is　simulated　to　control　in　the　continuous　stirred　tank　reactor.　The　results　indicate　that　the　proposed　control　strategy　has　an　excellent　system　performance　to　control　the　consistence　with　a　strong　anti-noise　performance.　Moreover,　the　system　based　on　the　improved　extreme　learning　machine　has　a　better　performance　than　that　based　on　extreme　learning　machine.