This paper proves in the context of maximum likelihood estimation of linear stochastic models of the Cowles Commission type [2], that if the model is fully identified and stable and the error variance matrix unrestricted, three-stage least-squares estimates differ asymptotically from full maximum likelihood estimates by order 1/T, where T is the number of time periods. When the full maximum likelihood estimates are best asymptotic normal so are the three-stage least-squares estimates.