Among the many methods currently available for estimating latent variable interactions, the unconstrained approach is attractive to applied researchers because of its relatively easy implementation with any structural equation modeling (SEM) software. Using a Monte Carlo simulation study, we extended and evaluated the unconstrained approach to estimate Level 2 latent interactions in multilevel SEM. We compared mean-centered and residual-centered methods for forming product indicators. We found that using the unconstrained approach with either centering method produced unbiased parameter estimates and standard errors of Level 2 latent interactions when intraclass correlations and group sizes are large. The accuracy of estimation of latent interactions depended on the reliability of the indicators of the Level 2 constructs. When the reliability of indicators was low, parameter estimates became unacceptably biased, but not standard errors. The residual-centered method was found to be more sensitive to decreases in reliability of indicators. (Contains 3 tables and 1 figure.)