After exposure to a novel food an association is thought to form between its sensory characteristics and its post-ingestive action. When the food is energy dense, its sensory properties can come to signal this fact and meal size is moderated accordingly. In humans, evidence for this ‘learned satiety’ is equivocal. However, previous studies have focused on measures of ad lib food consumption. Outside the laboratory we tend to make decisions about portion size and so energy intake is determined before a meal begins. Therefore, we reasoned that learning might be more likely to be expressed in expectations about the satiating quality of food (i.e., whether it is expected to be ‘filling’). Female participants (N=58) sampled a novel dessert and then provided a measure of their expectations relative to two very familiar foods. Half of the participants then consumed a low-energy dessert (228 kcal). The other half consumed a high-energy dessert (552 kcal). Both desserts were matched for their sensory properties. The next day expectations were assessed again and the participants were offered ad lib access to an intermediate energy-dense dessert. Although both groups consumed a similar amount of this dessert, ‘learned expected satiety’ was clearly evident—participants in the high-energy condition increased their expectations about the satiating quality of the dessert whereas those in the low-energy condition did not. This is the first evidence for learning of this kind and it challenges claims that humans are insensitive to covert manipulations to the energy density of food.