Previous research shows that one can attend effectively to one of multiple simultaneous voices, but switching attention between them incurs a substantial “switch cost”. Using a “voice switching” paradigm where the target voice was defined (and visually cued), by gender or location (target dimension separated by session), we investigated the effect of switching the irrelevant dimension. Feature binding accounts posit that features of the irrelevant perceptual dimensions of a stimulus might become bound to features of the relevant dimension. In particular, a change of feature of the irrelevant dimension, relative to its repetition, could reduce the benefit of attending to the same feature and/or facilitate switching to a different feature of the relevant dimension, due to automatic retrieval of the feature of the irrelevant dimension. We investigated this potential effect of binding on the switch cost (the cost of changing the feature of the relevant dimension) and its modulation by preparation (cue-stimulus) interval. The switch cost was reduced when the feature of the irrelevant dimension changed, mainly due to worse performance for feature repetitions on the relevant dimension; this was observed irrespectively of which dimension was relevant. A longer preparation interval significantly reduced the switch cost but did not interact with the effect of binding on the switch cost. This suggests two distinct types of process: top-down preparatory tuning of attention to a perceptual dimension, and bottom up, automatic, binding between features of the relevant and irrelevant dimensions, which is impervious to the effects of preparatory top-down control.