While primary sensory areas are robustly activated by sensory input signals, both theory and experiment suggest that the same cortices are modulated by internally generated activity. Predictive coding theories propose that sensory cortical responses incorporate comparisons between incoming sensory signals and expectations generated by internal models of the sensory scene. Yet several aspects of how sensory expectation is expressed in sensory circuits remain unclear. A fundamental question is whether distinct prediction effects accompany functional and anatomical differences between sensory modalities. The primary olfactory cortex (OC) receives unpatterned sensory inputs via the main olfactory bulb, and OC responses to odors reflect local associative dynamics more so than input activity. How sensory predictions are implemented in this system is not well understood. To address this, we chronically recorded neural populations in the mouse OC during presentation of odor pairs without reinforcement. 12 odors were systematically paired to establish stimulus-specific sensory predictions. In a subset of trials, the second odor was either presented alone (unpredicted) or omitted. After odor pairing, we observed changes to odor responses in both single cells and on the population level. Prediction-matched responses were bidirectionally modulated across the neural population, and became decodable from neural activity prior to odor onset. Surprisingly, after just one session, we observed reinstatement of the predicted odor representation instead of a mismatch response in trials where the predicted odor was omitted. These data indicate that prediction generates activity in the primary olfactory cortex that differs from what has been observed in other sensory modalities. This work was funded by an ERC advanced grant and an FCT SR&TD grant.