This paper compares the asymptotic efficiency of a number of two step estimators developed for estimating a static linear regression model with serially correlated errors when some observations are missing. A Monte Carlo simulation is used to illustrate the results in small samples.