In recent years, the simultaneous recording of large populations of cortical neurons have illuminated how sensory, motor, and cognitive processes affect the response variability of groups of neurons. Trial-averaged neuronal responses can be explained by a wide variety of cortical models; however, the circuit mechanisms that underlie how the shared neural variability (noise correlation) depends on cortical state are poorly understood. Visual attention, which improves perception of attended locations or features, is particularly well-suited to constrain models of cortical circuits because attention both increases the firing rate and stimulus sensitivity of neuron response, while simultaneously decreasing noise correlations. We provide a novel analysis of spike responses in visual area V4 that supports the hypothesis that a single biophysical mechanism underlies these diverse neural correlates of attentional modulation. Thus motivated, we develop rate and spiking models of recurrent cortical networks where top-down mediated increases in excitability, distributed across excitatory and inhibitory targets, capture the key neural correlates of attention observed in our data. Our models predict that top-down signals are principally directed to inhibitory neurons, whereas excitatory neurons receive more sharply tuned bottom-up inputs. These are two ideas with existing physiological support, yet they have not been derived as conditions for the neural correlates of attention. Our study provides an important step in linking the dynamics of recurrent cortical networks to the necessary structure of modulations that control their behavior. This is joint work with the laboratory of Marlene Cohen.