When testing means of multiple constructs measured using multiple items/variables, researchers often choose multivariate analysis of variance (MANOVA). An emerging argument against the universal use of MANOVA, based on development of the structural equation modeling (SEM), states that analysis methods should be determined by the nature of the construct (Cole, Maxwell, Arvey, & Salas, 1993). When the constructs are of an emergent variable system, measured items/variables are causal agents of the constructs (Bollen & Lennox, 1991). In this case, MANOVA is the best choice because group differences are examined on composites that maximally differentiate the groups on multivariate level. When the constructs are of a latent variable system, the measured items/variables are likely to co-vary due to the influence from an underline trait. In this case, MANOVA should not be used because of the potential measurement error in the composites and the Structural Mean Modeling (SMM) of the SEM should be used where the latent variables are theoretically error-free (Thompson & Green, 2006). After briefly describing characteristics, including advantages and disadvantages, of each method, we will illustrate these methods and their differences in results. The data employed were elementary school students in an experimental physical education curriculum group (n=1,749) and a comparison group (n=1,985). The latent variable system is a five-factor situational interest (SI) model. The latent factors are attention, challenge, exploration, novelty, and enjoyment, each was measured with 4 items. MANOVA yielded, with statistical significance (F=87.740, p<.001), that the experimental group rated challenge and exploration higher, but novelty, enjoyment, and attention demand lower than the comparison group. In the SMM, we first tested model-data fit by constraining corresponding loadings and intercept terms across groups to be identical. Results indicated a good fit (CFI=.985, SRMR=.043, RMSEA=.051), which ensured a plausible comparison of group means. With error term completely controlled, SMM produced results consistent with those of MANOVA on challenge (z=-6.637) and exploration (z=-12.588), the experimental group rated the two latent factors higher than the comparison. However, in contrast to the results from MANOVA, SMM indicated that experimental group students rated attention significantly higher (z=-7.798) than the comparison group students; and there was no statistically significant difference in enjoyment and novelty. With controlling possible errors in the composite scores, the SMM analysis suggests that MANOVA might have generated misleading results. In addition, SMM results seem to be consistent with both theoretical assumptions and data from other sources (e.g., direct observations, interviews).Keyword(s): curriculum, elementary issues, physical education PK-12