Perhaps the most successful marriage between neuroscience and computer science is the theory that dopamine cell firing encodes “temporal-difference reward prediction errors” (RPEs). RPEs are learning signals, that serve to update predictions of future reward and thereby guide adaptive decision-making. Yet this theory is clearly incomplete, since dopamine is not just a learning signal but is also critically involved in motivation. Boosting dopamine immediately enhances willingness to work at behavioral tasks, and dopamine release ramps up as animals get closer to reward, in a manner that resembles motivational value rather than RPE. I will discuss our recent results and hypotheses that seek to reconcile and integrate these learning and motivational functions of dopamine.