Attention enhances behavioral performance primarily by reducing interneuronal correlations, not by increasing firing rates. In visual area V4 of rhesus monkeys, attention improved population sensitivity by decreasing trial-to-trial variability between neurons, rather than by increasing individual neuron firing rates. This reduction in correlations between neurons' responses significantly improved the population's ability to detect stimulus changes, even though individual neurons showed only small changes in firing rates or noise. Attention reduced noise correlations between neurons, which is critical for population coding. The study shows that attentional modulation of noise correlations is the main factor driving improvements in population sensitivity. This finding suggests that understanding how attention affects neural populations requires considering interactions between neurons, not just individual responses. The results highlight the importance of pairwise correlations in population coding and indicate that future studies should focus on multi-electrode or imaging technologies to capture population dynamics on a behavioral timescale.Attention enhances behavioral performance primarily by reducing interneuronal correlations, not by increasing firing rates. In visual area V4 of rhesus monkeys, attention improved population sensitivity by decreasing trial-to-trial variability between neurons, rather than by increasing individual neuron firing rates. This reduction in correlations between neurons' responses significantly improved the population's ability to detect stimulus changes, even though individual neurons showed only small changes in firing rates or noise. Attention reduced noise correlations between neurons, which is critical for population coding. The study shows that attentional modulation of noise correlations is the main factor driving improvements in population sensitivity. This finding suggests that understanding how attention affects neural populations requires considering interactions between neurons, not just individual responses. The results highlight the importance of pairwise correlations in population coding and indicate that future studies should focus on multi-electrode or imaging technologies to capture population dynamics on a behavioral timescale.