Accumulating research indicates that physical activity is motivated by automatic evaluations of physical activity. Little is known about the stability of automatic evaluations or how their dynamics impact physical activity. We tested the measurement invariance and stability of university students’ (N = 164) automatic evaluations of physical activity. In addition, multiple regression and structural equation models with latent interaction variables were used to investigate how changes in automatic evaluations related to change in self-reported physical activity and differences in the level of directly measured physical activity. It was revealed that automatic evaluations had strict measurement invariance and that automatic evaluations have both stable and unstable components. People whose unfavorable automatic evaluations became more favorable over the week showed a larger increase in self-reported physical activity from the previous week than did people whose automatic evaluations remained unfavorable. These results indicated that the dynamics of automatic evaluations and physical activity can be intertwined.