In this study, guidelines for identifying the first-level error covariance structures in latent growth modeling (LGM) are proposed, assuming that the second-level error covariances are unstructured. The guidelines are based on the sequential chi-square difference test, adapted from Anderson and Gerbing (1988). Moreover, how to test for stationarity of an error process is specifically addressed. The guidelines are useful for correctly specifying the first-level error covariance structures, regardless of modeling growth curves for manifest variables or latent constructs. Simulated and real data were used to demonstrate identifying the first-level error covariance structures with the guidelines by using SAS PROC CALIS.