Localization accuracy for a wide Gaussian bar (sigma 1 degree) was essentially invariant with sampling density over a wide range. Luminance contrast, on the contrary, had a profound effect on localization accuracy. This difference suggests that the position is estimated from a few (3-4) samples from all those available in the image. A subsequent experiment confirmed that limiting the display to four samples did not impair localization accuracy. Computer simulations show that the result cannot be explained by the peak or centroid rule for position. The results imply that the visual system uses only a few samples to interpolate the luminance profiles, regardless of how many samples are available in the image.