Attention is a means for endogenously and selectively enhancing or suppressing the processing of sensory information to suit changing contexts. These changes are manifest with changes in the correlation structure of spiking activity within neuronal populations and changes in large-scale network oscillations reflected in local field potentials and the scalp EEG. However, the nature of the neural signals that drive changes in attention state are not understood. One hypothesis is that feedback signals from frontal cortical areas specifically target inhibitory subpopulations in sensory areas during attention, perhaps mediated by cholinergic pathways. This inhibitory drive is hypothesized in turn to bias the outcome of competitive interactions between excitatory subpopulations representing the attended or ignored attributes of stimuli. We predicted that if inhibitory neurons receive shared input during attentional selection it would lead to greater correlated variability in pairs of inhibitory neurons during the attended state compared to the correlated variability in pairs of excitatory neurons. Previously, correlated variability across the population as a whole has been shown to decrease with attention, so a relative increase for one specific subpopulation might not be expected, and would be consistent with the presence of a shared private signal to the inhibitory cells. We tested this prediction by recording neurons in V4 of macaque monkeys performing a selective spatial attention task. We classified these neurons as putative inhibitory (fast-spiking) or excitatory (regular-spiking) cells on the basis of three measures of their waveform shape known to differ between verified inhibitory and excitatory cells. We compared correlated variability in pairs of inhibitory neurons with that in pairs of excitatory neurons when attention was directed toward their receptive fields and away. We found that attention diminished correlations between excitatory cells, whereas correlations between inhibitory neurons did not decrease with attention, rather, the tendency was for inhibitory neurons to become more correlated with attention. This pattern of results is consistent with inhibitory neurons receiving a shared signal during selective attention.