The problem of identifying a restricted complexity model for control purposes is considered in this contribution. A new iterative identification method is proposed in which local and global information about the control design criterion is blended. The global information, which is used for faster convergence, comes from the usual extrapolating property of a model and the local information, used for higher accuracy, is the sensitivity of the closed loop system to the model parameters. It is shown that the method has the same stationary points as the control design criterion and it is also shown that if the approximated Hessian is sufficiently accurate, the method converge locally.