Abstract: The new identification method for nonlinear systems is presented, which combinesrecursive least-square (RLS) parameter estimation with nonlinear programming (BFGS).For this nonlinear system identification algorithm, the partial innovation of the measureddata is utilized by RLS, and another by BFGS. Because of nonlinearity in the system,the 'rough' parameter estimate is recursively RLS, and then the rough parameter estimateis iteratively 'refined' by BFGS. When this method is used for the identification oflinear systems, the accuracy of parameter estimate is enhanced and the convergence ofestimate is improved. It is the nonlinear iterative optimization in identification algori-thm that can overcome the nonlinear effect of process system identification, and improvethe accuracy and the convergence of parameter estimation. That has been fully verifiedby digital simulation.