Midbrain dopamine systems are crucially involved in motivational processes underlying the learning and execution of goal-directed behaviour. Dopamine neurons in monkeys are uniformly activated by unpredicted appetitive stimuli such as food and liquid rewards and conditioned, reward-predicting stimuli. By contrast, fully predicted stimuli are ineffective, and the omission of predicted reward depresses their activity. These characteristics follow associative-learning rules, suggesting that dopamine responses report an error in reward prediction. Accordingly, neural network models are efficiently trained using a dopamine-like reinforcement signal. However, it is unknown whether the responses to environmental stimuli concern specific motivational attributes or reflect more general stimulus salience. To resolve this, we have compared dopamine impulse responses to motivationally opposing appetitive and aversive stimuli. In contrast to appetitive events, primary and conditioned non-noxious aversive stimuli either failed to activate dopamine neurons or, in cases of close resemblance with appetitive stimuli, induced weaker responses than appetitive stimuli. Thus, dopamine neurons preferentially report environmental stimuli with appetitive rather than aversive motivational value.