It has proved surprisingly hard to generalize from the rapid advances in our understanding of fundamental cognitive mechanisms to people’s real-life behavior. This is in part because the impact such mechanisms have in real life depends on how they interact with other aspects of a person’s mental state and environmental circumstances. I will present recent work uncovering one such dynamical interaction, involving reward learning and mood: unexpected rewards affect mood and mood in turn affects responses to subsequent rewards. I will present behavioral and neuroimaging evidence for this two-way interaction, and develop a computational model that reveals its adaptive and maladaptive consequences: on one hand, it can ‘correct’ learning to account for global changes in the availability of reward in the environment; on the other hand, it might give rise to unstable oscillatory dynamics that result in emotional instability. I will introduce a novel smartphone-based experimental approach that allows directly studying how such real-life cognitive processes unfold over time.