Research on behavioural activation (BA) for depression assumes that increasing the rate of rewarding activities leads to reduced depressive mood. Although the efficacy of BA treatment has been convincingly demonstrated, assumptions referring to the mechanisms underlying this effect have not been tested decisively yet. The Behavioural Activation Scale for Depression and the Beck Depression Inventory II were administered over 8 weeks in 161 reliably diagnosed patients with unipolar depression who underwent BA treatment in groups. Time-lagged associations between behavioural activity on depression at subsequent assessment (1 week later) and vice versa were modelled (a) with multilevel models and (b) dynamic panel models that eliminate shared constant factors and allow for reverse causation (e.g., depression affecting subsequent activation in the model for activation on depression). Both activation and depression changed significantly (model-based within effect sizes for activation = .70 and for depression = −.75). Higher activation and lower depression predict each other over time (mixed-effects, time-lagged model), but this association disappears in dynamic causal models. Change patterns were only analysed within a time frame of 1 week; smaller time intervals were not investigated. There were no objective observational data of activity patterns. In the present study, increased behavioural activity and reduced depressive mood co-occur on the week scale. They predict each other across time, but this may be due to shared causes only.