Garner's speeded classification task has been used as an effective tool to probe holistic processing of object shape. This is achieved by comparing classification performance of a given object dimension between two experimental conditions. Worse performance in a "filtering" condition in which a second, irrelevant dimension of the same object varies on a trial-to-trial basis, compared to a "baseline" condition in which the irrelevant dimension is held constant, is labeled Garner interference, and indicates that the two dimensions are processed in a holistic manner. About a decade ago, we used Garner's task to provide evidence for different frames of processing mediating action and perception. Unlike perceptual estimations, visually guided grasping showed no Garner interference when subjects were asked to reach out and grasp an object along a given dimension. In other words, slower reaction times were observed in the filtering compared to the baseline condition only for perceptual estimates but not for grasping. In two experiments, we extend these findings to kinematic measures beyond simple reaction times. The results showed that Garner interference is also expressed in the variability of the response, with more variable within-subject performance in the filtering compared to the baseline condition for perceptual estimates but not for grasping. These findings provide converging evidence for the idea that, unlike perception, which processes objects holistically, visually guided action is performed in an analytic manner.