Two ambiguously rotating objects with shared rotational axes tend to be perceived as rotating in the same direction. Last year we reported that the coupling between two objects is sharply reduced as one of the objects is made unambiguous. This finding is a problem for explanations of perceptual coupling based on either bottom-up cooperativity between similar representations, or based on the influence of top-down processing imposing scene constraints. Our model of grouping proposes that cortical feedback acts to resolve a percept from ambiguous or competing information, and that (i) the strength of feedback is proportional to the ambiguity in the visual stimulus, and (ii) feedback acts globally, consistent with neuroanatomical evidence that cortical feedback connections are largely divergent and less topographically specific relative to feedforward projections. This implies that perceptual coupling is a side effect of global feedback from a model of the object or scene that is evoked to assign an interpretation to the objects. Coupling is low between ambiguous and unambiguous objects because an unambiguous object elicits little or no feedback. This model is consistent with evidence that cortical feedback serves to improve figure-ground discrimination in situations where the stimulus has low salience (Hupé et al., 1998), and that attention to a particular property or feature in one part of the visual field results in a global priming effect for that property (Treue & Trujillo, 1999). One prediction of this idea is that viewing a partially biased ambiguous object generates greater feedback when the object is perceived as rotating opposite to its bias because stronger feedback is required to select and stabilize the percept that has weaker evidence. Thus, when a partially ambiguous object is perceived as rotating against its bias, the degree of rotational coupling with an ambiguous object should be increased. This prediction confirmed in an experiment.