Increasing population level physical activity (PA) requires initiation of new PA among the sedentary and maintenance of PA once initiated. The PA intervention literature has tended to focus on initiation and its mediators more than on maintenance. Our work integrates previous findings on mediators of initiation with theory about maintenance-specific mediation to identify empirically viable mediators of PA maintenance. Adults (N=1049) 50–70 years old were randomly assigned to the phone-based KAM intervention or usual care. All participants self-reported their moderate and vigorous PA (MVPA) and on 6 mediators pre-randomization, and 6, 12 and 24 months later. The mediators assessed were exercise self-efficacy (SE), barriers, and enjoyment; incorporation of PA into the self-concept; and social support received from family and friends. A structural equation-model specifying 7 indirect effects of KAM on MVPA kcal/wk was an excellent fit, ÷2(141)=282.11, CFI=.98, RMSEA=.03. KAM participants expended about 190 more MVPA kcal/wk, âc =.105, z=2.90, p=.004. The total effect was comprised of the direct KAM-MVPA path, âc’ =.052, z=1.52, p=.13, and the significant total indirect KAM-MVPA effects, total â=.050, z=3.23, p<.001. The significance of all but one KAM-mediator path, and all mediator- MVPA paths, demonstrate that KAM positively impacted constructs predictive of MVPA. The one significant indirect effect, â =.009, z=1.99, p<.05, showed that KAM significantly increased SE, which was positively related to PA self-concept, which was itself predictive of higher MVPA. Although there was significant total mediation, only one indirect effect was significant by itself. SE was upstream to two endogenously mediated effects. Self-concept was the strongest predictor of PA and figural in two indirect effects. Adequate PA maintenance models will need to be more complex than initiation models, and should incorporate relationships among multiple mediators and bi-directional mediator-PA relationships.