One of the central tasks for all theories of perception is to answer the question why do things look as they do. In contemporary visual science there are three major approaches to this question. The first is the ecological approach which states that things look as they do because the optical stimulation is what it is: the structure of ambient light produces all the relevant visual information. The other two approaches, connectionist and cognitivist, claim that our visual world is not a simple reflection of external stimulation, but it is rather an inner construction created by the visual system itself. However, these two approaches offer different definitions of the origin and nature of perceptual construction. The connectionist approach is dominantly aimed to define neural mechanisms (so-called neural networks) responsible for the articulation and completion of poorly specified stimulus representations, while the cognitivist approach emphasises the role of higher cognitive processes, such as long-term memory, deductive reasoning and the like. In this paper the interpretative power of the above-mentioned approaches in the case of amodal completion phenomena is evaluated. Amodal completion is a large class of phenomena that shows that perceptual description of the external world can include something that is not given in optical stimuli. For example, most things around us are partially occluded by other things, that is, they are not seen as 2-D mosaics of visible fragments, but rather as complete and continuous objects arranged in depth (one behind another). Referring to this, the simple question for three approaches is asked: how does our visual system complete the invisible parts of external objects. The analysis suggests the following: (1) The ecological approach generally fails to explain the fact that optical stimulation is not sufficient to describe amodal contours and surfaces. However, it points out that the stimulus constraints strongly influence what will be seen in any given optical scene. (2) The connectionist approach offers the most developed models of amodal completion, but it fails to explain some facts such as multistability of amodal completion (e.g. two or three equally probable perceptual descriptions of the same stimulus pattern). (3) Finally, the cognitivist concepts, such as past experience and familiarity, fail in prediction.