The cortex has more “backward” projections than it does “forward”, which fits well to early hypotheses that the brain uses memory and expectation to form predictions of input. Our lab has established a way to measure the role of feedback in prediction, using functional magnetic resonance imaging combined with retinotopic mapping to identify regions of V1 receiving no sensory stimulation (and therefore only feedback/lateral input). We use multivariate pattern classification to decode the information content of these top-down signals. We are now extending this methodology to high resolution laminar analysis using 7Tesla imaging. In a seminal finding, we again blocked feedforward input to a region of V1 and measured contextual feedback, but now using 7T brain imaging (0.8 mm3). We found that contextual feedback information peaks in superficial layers of V1. We also observed that contextual feedback signals to V1 have a spatial imprecision compared to feedforward inputs. Left: Non-stimulated V1 receives cortical feedback information to superficial layers that is predictive of visual context