The spatial focus of attention has traditionally been envisioned as a simple spatial gradient of enhanced activity that falls off monotonically with increasing distance. Here, we show with highdensity magnetoencephalographic recordings in human observers that the focus of attention is not a simple monotonic gradient but instead contains an excitatory peak surrounded by a narrow inhibitory region. To demonstrate this center-surround profile, we asked subjects to focus attention onto a color pop-out target and then presented probe stimuli at various distances from the target. We observed that the electromagnetic response to the probe was enhanced when the probe was presented at the location of the target, but the probe response was suppressed in a narrow zone surrounding the target and then recovered at more distant locations. Withdrawing attention from the pop-out target by engaging observers in a demanding foveal task eliminated this pattern, confirming a truly attention-driven effect. These results indicate that neural enhancement and suppression coexist in a spatially structured manner that is optimal to attenuate the most deleterious noise during visual object identification.attention ͉ magnetoencephalography ͉ visual I t is a common experience that one is able to focus on relevant parts of a visual scene even when irrelevant parts of the scene are more salient. It has been suggested that this voluntary focusing is mediated by a biasing of competitive stimulus interactions in the visual cortex, which promotes preferential processing of relevant over irrelevant input (1-3). This competitive advantage can be achieved by enhancing the processing of relevant inputs or by attenuating the processing of irrelevant inputs. Evidence has accumulated that attention operates by means of both neural enhancement (4-7) and neural suppression (8-13). More recent data from functional MRI in humans indicate that enhancement and suppression may cooperate across the visual scene (14, 15), leading to an increase in selectivity in a push-pull-like manner (15). That is, a spatially organized combination of enhancement and suppression may effectively sharpen the demarcation of relevant from irrelevant inputs, particularly in cluttered visual scenes in which neural representations of relevant and irrelevant information may become mixed together (16, 17) .Many behavioral and electrophysiological studies of attention have indicated that attending to a location produces a monotonic gradient of processing efficiency around the attended location (18)(19)(20)(21)(22). In contrast, computational models motivated by the known anatomy and physiology of the primate visual system have predicted that the spatial distribution of cortical activity around the focus of attention may be more complex than a simple gradient. In particular, the selective tuning model (ST) of Tsotsos et al. (23,24) proposes an architecture of attentional selection that explicitly predicts a suppressive zone surrounding the focus of attention. In short, ST provides an acc...