Appropriate arousal in our brains critically relies on hypothalamic hypocretin/orexin neurons (HONs). HON activity regulation has been a subject of intense research, uncovering slow responses to nutrients and hormones, and fast responses to external sensory inputs. However, it remains unclear whether HONs also respond to statistical patterns in the environment. Among these, reward probability and uncertainty are fundamental measures of information that enable evaluation of predictions, and guide diverse aspects of cognition such as learning and motivation. Using different cues to denote reward likelihood across the full probability range (0-1) , we found that HON population responses to the reward-predicting cues co-vary with reward uncertainty. Single-cell analyses additionally revealed HON subsets specializing in monotonic representations of probability, as well as cells sharply tuned to uncertainty. Furthermore, simultaneous recordings of HONs and midbrain dopamine neurons (DANs) indicated that HONs and DANs diverge in their coding of reward probability, both in responses to reward-predicting cues and to rewards. These results demonstrate previously unseen coding of reward expectation in genetically-defined hypothalamic cells central to clinical conditions and neurobehavioral physiology, and suggest a possible role for HONs in converting reward statistics into arousal-based allocation of cognitive and autonomic resources.