Although inhibition of return (IOR) is widely believed to aid search by discouraging reexamination of previously inspected locations, its impact actually appears to decline as the number of target locations increases. We test three possible reasons for this paradoxical result: (1) IOR is capacity-limited, (2) IOR is sensitive to subtle changes in target location probability, and (3) IOR decays with distance from a previously attended location. The present investigation provides strong support for the third explanation, indicating that a gradient of inhibition is centered on previously attended locations. We note that this inhibitory gradient resolves a paradox in the literature. Moreover, we speculate that the inhibitory gradient may reflect a "similarity space" within which target locations near to the cue are tagged with inhibition due to their similarity to the cued location. The farther the target location is away, the less similar it is to the cued location, and thus the less inhibition it receives.