While the effect of selection in predictive validity studies has long been recognized and discussed in psychometric studies, little consideration has been given to this problem in the context of latent variable models. In a recent paper, Muthén&Hsu (1993) proposed and compared estimators of predictive validity of a multifactorial test. Both selectivity and measurement error were considered in the estimation of predictive validity. The purpose of the present paper is to expand on Muthén&Hsu (1993) by examining and comparing the sampling behaviour of three estimators for predictive validity, LQL (listwise, quasi‐likelihood estimator), FQL (full, quasi‐likelihood estimator) and FS (factor score estimator), using a Monte Carlo approach. Effects of selection procedures, selection ratios and sample sizes on the sampling behaviours of the estimators are also investigated. The results show that FQL and FS are the two preferred estimators and each has different strengths and weaknesses. A real data application is presented to illustrate the practical implementation of the estim