What we expect to see can greatly affect what we perceive. How does expectation influence local circuitry among visual cortical neurons to allow for perceptual discrimination in the context of rich natural inputs? Little work has sought to understand how populations of sensory cortical neurons change their responses with expectation, and the majority of studies manipulate expectation in a task-irrelevant way, disallowing for comparisons of changes in neural activity with behavioral performance. Theoretical and experimental work indicates that changes at the population level, in particular those related to the variability in neural responses, greatly impact stimulus encoding. Therefore, we sought to investigate whether changes in the structure of neural population activity in mid-level visual cortex might underlie the behavioral advantages conferred by forming an expectation. We recorded from populations of visual cortical area V4 neurons using implanted 96-electrode arrays in macaques engaged in a natural scene change detection task in which we modulated image expectation. During high expectation blocks, the same image (e.g. image A) was used on every trial. During low expectation blocks, a random image (from 10 possible images, including image A) was chosen for each trial, thereby reducing the expectation that image A would be selected for a given trial. Comparisons were made for image A during high and low expectation blocks. Data from 2 monkeys showed a robust improvement in behavioral performance when the image was expected. Our recordings showed that expectation decreased neural responses and modulated noise correlations. Using dimensionality reduction methods, we found a decrease in shared variability and increase in dimensionality with high expectation. Our results support the idea that expectation is built through the interactions among populations of neurons relatively early in the visual system, enabling it to be flexible for arbitrary visual scenes and objects.