This study investigated the relationship between motivation instability and its predictive factors in synchronous online classes. Perception of class content (perceived interest, difficulty, and utility), sleepiness/fatigue, and learning anxiety were predictive factors. Data from 78 and 72 students, respectively, were collected from an online survey conducted at the beginning (Group 1) and end of the semester (Group 2) at University A, and data from 50 students were collected from an online survey conducted at University B (Group 3). A multi-group analysis of the three groups was conducted using structural equation modeling. Our model constructed equality constraints between Groups 1 and 2 for all path coefficients, while those in Group 3 were freely estimated. The results showed that sleepiness/fatigue, learning anxiety, and perceived difficulty of class content correlated positively with motivation instability. However, the relationship between motivation instability and its predictive factors varied between groups.