Objective: To investigate how the effects of a group-based intervention program (MoVo-LISA) on exercise behaviour were mediated by cognitive variables. Different causal models mapping the short-term (adoption) and long-term (maintenance) intervention effects were tested using path analyses.

Design: N = 220 in-patients of a rehabilitation clinic were assigned to an usual care or intervention group (quasi-experimental design). Questionnaire-based assessment was conducted at baseline; discharge; and at six weeks, six months and 12 months post discharge.

Measures: The potential mediator variables were outcome expectations, self-efficacy, strength of goal intention (intention strength), self-concordance, action planning and barrier management.

Results: Observed intervention effects on exercise behaviour (p < 0.05) were mediated by intention strength at the adoption and maintenance stages, by action planning only at the adoption, and by barrier management only at the maintenance stage. Self-efficacy and outcome expectations were only indirectly involved in these mediations by affecting intention strength and self-concordance.

Conclusion: This is the first study to track the cognitive mediation processes of intervention effects on exercise behaviour over a long time-period by differentiating the adoption and maintenance stages of behaviour change. The findings emphasise the importance of deconstructing intervention effects (modifiability vs. predictive power of a mediator) to develop more effective interventions.