Interest operators find "interesting" locations in images. Existing algorithms are based on heuristic measures, such as local contrast, second order structure, or entropy. Here, we learn an interest operator from human eye movements. This is both benefical in computer vision applications, and it provides insight into the human eye movement system.