<p>We have previously demonstrated that gaze is focused on areas of faces characterized by the greatest increase in total contrast. Total contrast is defined as the sum of the contrasts of all luminance gradients in the image region being analyzed, and increases in this indicator are taken as the difference from the total contrast of the surrounding region. To demonstrate that these regions are determined preattentively as attentional targets, the present study used stimuli in which the regions of interest could not be predetermined endogenously on the basis of the semantic properties of the image. Subjects were presented with fragments of natural scenes. For each stimulus, averaged fixation density distribution maps were constructed. Images were processed using a program determining the distribution of total contrast across the image. For each spatial frequency, two 2D maps of the locations of regions characterized by the greatest (max) and smallest (min) increases in total contrast were calculated. Empirical maps were compared with the calculated min and max maps using two complementary metrics: CC and NSS. Similarity between the calculated and empirical maps was found to increase significantly with increases in the difference in the total contrast between software-selected areas and their surroundings. Max maps were then compared with random (ran) maps obtained by randomly placing “regions of interest” within the images. Again, similarity of the empirical maps to the max maps was significantly greater than similarity to the ran maps. These results lead to the conclusion that the targets of exogenous attention are the image areas with the greatest increase in total contrast. These areas of perceptual salience are automatically extracted from the input image in different spatial-frequency ranges by second-order visual filters which combine information about local luminence gradients.</p>