In this paper, we show that salient regions, detected by a biologically motivated attention system, have a considerably higher repeatability than features detected by standard detectors (Harris-Laplacians and DoG features). If a scene contains a salient region, which differs in one of several possible features from the rest of the scene, the repeatability is close to 100%. In settings in which no especially salient region is present, the repeatability of attention regions is similar to DoG features and considerably higher than Harris-Laplacians.