This paper presents a gradient-based iterative identification algorithms for Box_Jenkins systems with finite measurement input/output data. Compared with the pseudo-linear regression stochastic gradient approach, the proposed algorithm updates the pa