School engagement as a multidimensional construct has been known not only as a predictor of academic achievement, school adjustment, and school dropout but also as a mediator in the relations between school/family context or personal variables and adjustment outcomes. This study was conducted to test the fit of a multidimensional model of school engagement consisting of three factors (i.e., affective, behavioral, cognitive) and examine the decreasing tendency of school engagement during the transition to middle school. Wang, Willett, and Eccles’s (2011) school engagement measures were administered to 1,172 students in Grades 5 to 8. Confirmatory factor analyses and latent mean analyses were conducted to examine research questions. First, confirmatory factor analyses revealed that, in all of the four groups (i.e., elementary school boys, elementary school girls, middle school boys, middle school girls), the three-factor model of school engagement fit the data better than the one-factor model and the two-factor model. Second, the four groups did not significantly differ in the composition of the three factors and factor loadings. Third, there were no significant differences in the intercepts of the observed variables between elementary and middle school students in each gender. Given that both factor loadings and intercepts were invariant, latent mean analyses were conducted to test the latent mean differences on affective, behavioral, and cognitive engagement. Both boys and girls showed a significant decrease in each factor of school engagement. The effect sizes of latent mean differences on each factor by gender and school level showed that after the transition to middle school affective engagement declined to a greater extent than did behavioral and cognitive engagement and the three factors of school engagement decreased to a greater extent for girls than for boys. Finally, the necessity of identifying school engagement styles and testing the second-order model with one second-order factor (i.e., school engagement as a meta construct) and several first-order factors (e.g., affective, behavioral, cognitive engagement) was discussed.