Dopamine projections from the midbrain to the striatum and frontal cortex are involved in behavioral reactions controlled by rewards, as inferred from deficits in parkinsonism, schizophrenia, and drug addiction. Recent experiments have shown that dopamine neurons are not directly modulated in relation to movements. Rather, they appear to code the rewarding aspects of environmental stimuli. They show short, phasic increases of activity following primary food and liquid rewards ("unconditioned stimuli") and conditioned, reward-predicting stimuli of visual, auditory, and somatosensory modalities. They also display smaller activation-depression sequences after stimuli resembling rewards and after novel or particularly intense stimuli. Rewards are only reported as far as they occur differently than predicted. According to learning theories, a "prediction error" message may constitute a powerful teaching signal for behavior and learning. The phasic reward message is different from the more tonic enabling function of dopamine that is deficient in Parkinson's disease, indicating that dopamine neurons subserve different functions at different time scales. Neurons in other brain structures, such as the striatum, orbitofrontal cortex, and amygdala, code the quality, quantity, and preference of rewards. The dopamine reward prediction error signal may cooperate with these reward perception signals during the learning and performance of behavioral reactions to motivating environmental stimuli.