Abstract
An asymptotically distributions-free method for structural equation models with presence of missing observations is developed via the generalized least squares approach. Asymptotic properties of the estimator are studied and an iterative procedure is presented to produce the estimates, standard error estimates and the goodness of-fit statistic. Under the elliptical distribution assumption. a computer program is implemented in t h e context of a confirmatory factor analysis model. Numerical results that based on a longitudinal study are reported.