Multiple dots moving independently back and forth on a flat screen induce a compelling illusion of a sphere rotating in depth (structure-from-motion). If all dots simultaneously reverse their direction of motion, two perceptual outcomes are possible: either the illusory rotation reverses as well (and the illusory depth of each dot is maintained), or the illusory rotation is maintained (but the illusory depth of each dot reverses). We investigated the role of attention in these ambiguous reversals. Greater availability of attention – as manipulated with a concurrent task or inferred from eye movement statistics – shifted the balance in favor of reversing illusory rotation (rather than depth). On the other hand, volitional control over illusory reversals was limited and did not depend on tracking individual dots during the direction reversal. Finally, display properties strongly influenced ambiguous reversals. Any asymmetries between ‘front’ and ‘back’ surfaces – created either on purpose by coloring or accidentally by random dot placement – also shifted the balance in favor of reversing illusory rotation (rather than depth). We conclude that the outcome of ambiguous reversals depends on attention, specifically on attention to the illusory sphere and its surface irregularities, but not on attentive tracking of individual surface dots.

We compared the spatial and temporal allocation of attention as revealed by microsaccades. Observers viewed several concurrent "rapid serial visual presentation" (RSVP) streams in the periphery while maintaining fixation. They continually attended to, and discriminated targets in one particular, cued stream. Over and above this continuous allocation, spatial attention transients ("attention shifts") were prompted by changes in the cued stream location and temporal attention transients ("attentional blinks") by successful target discriminations. Note that the RSVP paradigm avoided the preparatory suppression of microsaccades in anticipation of stimulus or task events, which had been prominent in earlier studies. Both stream changes and target discriminations evoked residual modulations of microsaccade rate and direction, which were consistent with the presumed attentional dynamics in each case (i.e., attention shift and attentional blink, respectively). Interestingly, even microsaccades associated with neither stream change nor target discrimination reflected the continuous allocation of attention, inasmuch as their direction was aligned with the meridian of the target stream. We conclude that attentional allocation shapes microsaccadic activity continuously, not merely during dynamic episodes such as attentional shifts or blinks.