Technological and scientific images, and other images with epistemic uses, have varied appearances and functions. They seem to be analog or symbolic representations available to researchers for a variety of epistemic purposes such as summarizing data, or presenting, discussing and verifying hypothetic propositions about the world. This article studies the perception and understanding of scientific/epistemic images within a conceptual framework grounded in the notion of reference. It introduces the hypothesis stating that the performance of the perceptual understanding of a particular scientific image depends on the epistemic uses of attention. The hypothesis suggests that understanding a scientific picture requires making an epistemic use of the attentional control of visual routines in order to obtain knowledge on the spatial structure and the referents of a particular image or graphic representation.