Difficulty of identification of moving average models lie in that unknown noise terms appear in the information vector. By means of the interactive estimation theory in hierarchicalidentification-using the estimation residual to replace those noise terms and of the multi-innovation identification theory-expanding the innovation length and making sufficient use of system observation data,a muhi-innovation-recursive least squares and a least-squares iterative algorithms are presented. Compared with the recursive extended least squares algorithms, the proposed two algorithms have fast convergence rates and can produce highly accurate parameter estimation. The simulation results indicate that the proposed alorithms have good performance.