Many theories have been offered to explain how the visual system registers, encodes, and recognizes the shape of an object. Some of the most influential assume that border lines and edges activate neurons in primary visual cortex, and these neurons encode the orientation, curvature, and linear extent of the shape as elemental cues. The present work challenges that assumption by showing that well-spaced dots can serve as effective shape cues. The experimental tasks drew on an inventory of unknown two-dimensional shapes, each being constructed as dots that marked the outer boundary, like an outline contour. A given shape was randomly picked from the inventory and was displayed only once as a target. The target shape was quickly followed by a low-density comparison shape that was derived from the target (matching) or from a different shape (non-matching). The respondent’s task was to provide a matching judgment, i.e. , deciding whether the comparison shape was the “same” or “different.” Clear evidence of non-chance decisions was found even when the matching shapes displayed only 5% of the number of dots in the target shapes. Visual encoding mechanisms allow a shape to be identified when it is displayed at various locations on the retina, or with rotation or changes in size. A number of hierarchical network (connectionist) models have been developed to accomplish this encoding step, and these models appear especially credible because they are inspired by the anatomy and physiology of the visual system. The present work demonstrates above-chance translation, rotation, and size invariance for unknown shapes that were seen only once. This is clearly at odds with connectionist models that require extensive training before a shape can be identified irrespective of location, rotation, or size.