There are many conditions in which the visually perceived shape of an object differs from its true shape. We here show that one can reveal such errors by studying grasping. Nine subjects were asked to grasp and lift elliptical cylinders that were placed vertically at eye height. We varied the cylinder's aspect ratios, orientations about the vertical axis and distances from the subject. We found that the subjects' grip orientations deviated systematically from the orientations that would give the mechanically optimal grip. That this is largely due to misjudging the cylinder's shape (rather than to selecting a comfortable posture) follows from the fact that the grip aperture was initially more strongly correlated with the maximal grip aperture (which is related to the expected contact positions) than with the final grip aperture (which is determined by the real contact positions). The correlation with the maximal grip aperture drops from 0.8 to 0.6 in the last 1% of the traversed distance (11% of movement time), showing that the grip aperture was anticipated incorrectly (it is automatically "corrected" at contact). The grip orientation was already strongly correlated with the grip orientation at the time of maximal grip aperture, half way through the movement (R > or = 0.7), showing that the suboptimal grip orientations were planned that way. We conclude that subjects plan their grasps using information that is based on the misperceived shape.