For multiple-input single-output output-error systems,a difficulty is that the information vector in the identification model contains the unknown system outputs (true outputs or noise-free outputs),thus the standard identification algorithm cannot be applied directly.This paper presents a stochastic gradient identification algorithm based on the unknown output estimation.The basic idea is to replace the true output with the output estimate which is predicted/estimated by the estimated model,and also to compute the system parameter estimates by using the output estimates.Convergence of the proposed algorithm is studied and a simulation example is provided.