where S is the empirical covariance matrix, the last two terms canbe omitted and a“quick”necessary condition for identification is d. Maximization of (8) is performed numerically in many commonly available software programs like AMOS, LISREL, Mplus etc. There are many situations where using this covariance based approach will encounter serious difficulties“for many complicated situations: for example, when deriving the covariance structure is difficult, or the data structures are complex”(Lee&Song, 2012, p. 15). Our goal here is to elaborate on the Bayesian estimation of SEM, illustrating its advantages and its reliability in small samples. We also present several complicated data generating processes or models where the Bayesian approach presents some unique advantages.