Several related estimators of the parameters of a single equation of a simultaneous equation model are proposed. They are shown to be consistent and asymptotically efficient when the residual of the equation follows the first-order autoregressive process. One of these estimators, called MS2SLS, is designed to be consistent under a more general assumption about the stochastic process of the residual. A numerical analysis shows that the efficiency of MS2SLS is much higher than that of 2SLS under a general assumption about the stochastic process of the residual.