The human visual system has the ability to correctly interpret most images it receives from the environment. Underlying this ability are sophisticated perceptual grouping mechanisms that are able to link together image features that arise from the same physical source (e.g., the same object). Object recognition, and many other perceptual capabilities, would be impossible without appropriate perceptual grouping. We have proposed a theoretical framework for perceptual grouping (Geisler & Super, 2000) and have been using that framework as a starting point for research on specific perceptual grouping mechanisms. Recently, contour grouping has been of particular interest in the lab (Geisler, Perry, Super & Gallogly, 2001). In a typical contour grouping task the observer is required to detect a line-segment contour in a background of randomly oriented and positioned line segments. One such display is illustrated in A. We have developed a simple one-parameter model for contour grouping that consists of three processing stages: (1) extraction of local contour elements using spatial filtering like that observed in primary visual cortex, (2) formation of binding strengths between local contour elements according to a local grouping function derived from the statistics of natural images, and (3) integration of local elements into global contours based upon a transitive grouping rule: if element a binds to b and b binds to c then a becomes bound to c. The solid lines in B show the groups found by the model for the pattern in A. We have found that this model does a good job of quantitatively predicting human contour detection performance under a wide range of conditions.