Previous work has shown that HJA, a patient suffering from visual agnosia, can complete occluded contours whilst being impaired at assigning contours to foreground and background figures (Giersch, Humphreys, Boucart, & Kovacs, 2000). Here we tested whether completed contours are automatically bound with visible contours, after being derived from them. HJA, lorazepam-treated and nontreated healthy participants were asked to match a first reference line with an equal or longer line of identical orientation included in one of two lateral figures. The target line was in the foreground or the background of the figures. The distractor picture included two short collinear line-segments belonging to two different figures, so that participants had to process the occluded parts to discriminate the target from the distractor line. When the target line was in the background, both HJA and lorazepam-treated participants were faster when the length of the reference line corresponded to the length of the occluded part of the target line, relative to when it corresponded to the length of the occluded part plus a visible contour. In contrast, control participants tended to show an advantage for matching a reference line whose length was the same as the visible contours plus the occluded part. However, when the stimuli were displayed for 50 ms only and then masked, controls showed the same results as HJA. These results suggest that responses in the matching tasks are biased by the existence of an early completed occluded line that remains isolated from real contours.