The purpose of this study was to explore the longitudinal structural and measurement invariance of the 17-item Philadelphia Geriatric Center Morale Scale (PGCMS) using a multidimensional item response model and the unidimensional Rasch model in contrast to prior work employing classical test theory. Participants (n = 177) were assigned to three interventions and assessed at three time points. The structural analyses found that a 3-factor model fit better than a 1-factor or a 2-factor model using the Akaike information criterion (AIC) and Schwartz’s Bayesian information criterion (BIC) though not with the AICPC or the BIC PC. With respect to item fit, differential item functioning (DIF) and correlation of item logit positions, the PGCMS was quite stable. However, Items 4 and 5 were problematic in terms of fit. The internal consistency reliabilities of three factors were also relatively lower than desirable at one point and adequate at others.