Recent evidence challenges the view that attention acts on the outputs of e
arly filters dedicated to processing motion, color, and orientation. Instea
d, "proto-objects" specified by shading, depth, direction of lighting, and
surface information are thought to provide input to attentional processing.
These findings are extended here to the parsing of occlusion-based contour
s. Multicolored occlusion structures were briefly presented and illusory co
njunctions measured. More illusory conjunctions were made to structures in
which color was inconsistent with form information, a result that can be ex
plained by a property of the visual system that biases the integration of c
olor to be consistent with form. Results show that this constraint was base
d on global structural descriptions rather than the local information provi
ded by T-junctions and collinearity. Together, these results offer a new to
ol for the study of the binding problem in vision.