In this paper we investigate the impact of misspecification of disturbance in the white noise term on the GARCH processing. We consider an AR (1)-GARCH (1, 1) model and compare the performance of parameters estimation with two optimization algorithms. The results show that misspecification of disturbance has little effect on the parameters estimation of conditional mean equation, but it will lead to a big biased value in conditional variance equation when we check the coefficients of ARCH and GARCH terms. Moreover, in the case of small sample, the NLMINB algorithm reaches a better estimator for the coefficients of GARCH term than that of LBFGSB. On the other hand, as the sample size increases, the LBFGSB algorithm will demonstrate faster convergent rate than that of NLMINB algorithm.