HOW does the human visual system ’bind’ different fragments in the visual scene to create enduring representations of objects(1-5)? A visual illusion known as ’metacontrast’(6-9) or backward masking provides compelling evidence that perception is not instantaneous and that it occurs sequentially in distinct stages, If a solid white target square is displayed for 50 ms in a tachistoscope, switched off, and followed by a 50 ms display of two flanking mask squares, remarkably, subjects report seeing only the two flanking squares : the first square is simply not ’seen’. By plotting the magnitude of masking as a function of the delay between the target and mask (the stimulus onset asynchrony), one can obtain a characteristic ’U’-shaped function(7) with optimum masking occurring at about 50 ms, and no masking with synchronous target and mask presentations or at delays higher than 300 ms. The illusion is also highly sensitive to elementary stimulus dimensions such as colour, orientation and spatial frequency(8), and it has been suggested(10) that it is based on ’low level’ autonomous visual mechanisms rather than cognitive processes, Here we describe a novel visual stimulus that demonstrates that metacontrast can be strongly modulated by ’top down’ influences such as voluntary visual attention.