Results from a number of paradigms (including change blindness, inattentional blindness, integration over saccades, and backward masking) suggest that most of the visual information we take in is not retained, even for very short periods of time. This has led some to question whether such information is ever really perceived. We examine this issue using a variant of the classic metacontrast stimulus. When a briefly presented disk is followed by a briefly presented ring, observers may report not seeing the disk. Rather they report seeing the ring flicker as if the change in form from disk to ring is not recorded. This effect is highly dependent on the interval between the onset of the disk and the onset of the ring (the "stimulus onset asynchrony" or SOA). The maximum effect is usually found at a critical SOA of about 50 msec. Here we show that the ability of observers to distinguish such a disk/ring pair from a flickering ring is dependent also on how soon after the stimulus they respond. Early responses show a much smaller masking effect than late responses: Near the critical SOA accuracy improves when the observer responds more quickly (the opposite of the standard speed-accuracy trade-off), although at longer and shorter SOAs observers are less accurate on these early responses (a typical speed-accuracy trade-off). We interpret this finding as demonstrating that, at least in the case of metacontrast, retention of form information is disrupted, rather than initial access.