For nearly half a century now, Garner interference has been serving as the gold standard measure of dimensional interaction and selective attention. But the mechanisms that generate Garner interference are still not well understood. The current study proposes a novel theory that ascribes the interference (and dimensional interaction in general) to episodic feature integration processes at the micro (trial-to-trial) level. The novel account builds on earlier well-established notions of "feature integration" and "object files," and is augmented by formal derivations. The sequential binding account predicts that the magnitude of Garner interference is related to the strength of feature integration along consecutive trials. Three experiments were set to test this novel binding theory. Experiment 1 and Experiment 2 tested performance with integral dimensions (chroma and value, and width and height of rectangles); whereas Experiment 3 examined performance with a pair of separable dimensions (circle's size and angle of a diameter). In addition, the time lag ensuing between consecutive trials was manipulated. The results strongly supported the predictions of the sequential binding account: (a) with integral dimensions, substantial amounts of Garner interference were correlated with large partial repetition costs (e.g., consensual markers of feature integration), but this pattern was not observed with separable dimensions, and (b) the magnitude of both Garner interference and partial repetition costs diminished as a function of the ensuing time lag between consecutive trials, pointing to a common time-dependent memory mechanism. These results adduce strong support in the predictions of the feature binding theory of Garner interference, giving currency to the idea that dimensional interaction is driven by feature integration.