The role of feedback in sensory systems remains a major puzzle in neuroscience. Classical computational theories such as hierarchical predictive coding[1] or hierarchical Bayesian inference[2] postulate abstract normative goals for the role of feedback connections. However, predictions derived from these theories are typically specified at the level of individual neurons. Whether feedback shapes collective properties of neural populations and how such changes may impact encoding of sensory signals remain unknown. Here we tackle these questions directly by studying the population coding of natural stimuli in the dorsolateral lateral geniculate nucleus (dLGN) of behaving mice[3]. Extracellular electrophysiological recordings were performed in the mouse dLGN, while corticothalamic (CT) feedback from primary visual cortex (V1) was optogenetically suppressed. Our analysis was focused on two aims: i) determining whether CT feedback changes the structure of population activity as opposed to only modulating each neuron independently, and ii) exploring how these changes influence sensory coding. First, by estimating high-order correlations among neurons using multi-information, and by comparing distributions of population synchrony to carefully constructed null distributions, we found that feedback significantly increases network correlations. When feedback is suppressed, correlations among neurons tend to weaken and become more unstable. Second, information theoretic analysis revealed that these population-level changes exert a significant impact on sensory coding. By estimating response entropy we found that when feedback was intact, populations increased the variability of their responses to each movie segment. This variability was however not random - it enabled the population to use highly informative states more frequently, as quantified by symbol-specific information[4]. This coding strategy increased the overall mutual information between stimulus and population activity. Interestingly, our findings can not be immediately reconciled with the existing theories of feedback in sensory coding, which largely focus on individual neurons. Our results suggest therefore a need for expanding the theoretical understanding of the role of feedback modulation. Despite this discrepancy, we demonstrate that cortical feedback does play an important role in shaping efficient and informative population codes in lower levels of the sensory hierarchy.