We tested whether shape recognition computations contribute to the initial determination of figure and ground by asking observers to report about which region of briefly exposed masked stimuli appeared to be figure. The region on one side of a figure-ground contour always denoted a meaningful shape. In full versions of the stimuli, Gestalt variables favored the other, less denotative, region; in half versions of the stimuli, Gestalt variables were absent. If shape recognition inputs do not contribute to figure-ground computations, the less denotative region should be seen as figure in the full stimuli, and both regions should be seen as figure equally often in the half stimuli. Results show otherwise, suggesting that shape recognition routines do contribute to the initial determination of figure and ground, provided they have been completed, which seems to require approximately 150 msec for canonically oriented stimuli.