Studies that investigated stability of PA in older populations are scarce. Moreover, no studies used dynamic indicators to predict PA trajectories. The purpose of the present study were to investigate PA stability overtime, and to examine if changes in self-reported physical function (dynamic indicator) are better predictors of trajectories of PA than baseline measures of physical function (static indicator). This is a prospective postal survey with two time-point follow-ups: 12 and 38 months. Participants were older adults aged ≥60 years, and members of the medical insurance scheme of the French national education system. They responded to a self-report questionnaire on PA and general health status at three different times: baseline, 12- and 38-month follow-ups (n=243 for the 12-month follow-up; n=164 for the 38-month follow-up). Overtime analyses of PA showed a moderate-to-good stability with regard to both duration and volume of PA; however, a decrease in stability for vigorous PA was found between 12- and 38-month follow-ups. Both baseline measure and changes in physical function predicted PA trajectories, but magnitudes of associations were stronger for the dynamic indicator. Moreover, change in physical function was the only predictor of both becoming active compared with Inactive (reduced probability) and becoming inactive compared to Active (increased probability). In conclusion, a dynamic indicator of physical function is a better predictor of PA variation than static indicators.