In search arrays where the target is presented with similar nontarget stimuli, it is advantageous to shift the internal representation of the target features away from the nontarget features. According to optimal tuning theory (Navalpakkam & Itti, 2007), the shift of the attentional template increases the signal-to-noise ratio because the overlap of neural populations representing the target and nontarget features is reduced. While previous research has shown that the internal representation of the target is indeed shifted, there is little evidence in favor of a shift in attentional selectivity. To fill this gap, we used a cue-target paradigm where shorter reaction times (RTs) at cued than at uncued locations indicate attentional capture by the cue. Consistent with previous research, we found that attentional capture decreased with decreasing similarity between cue and target color. Importantly, target-similar cue colors closer to the nontarget colors captured attention less than target-similar cue colors further away from the nontarget colors, suggesting that attentional selectivity was biased away from the nontarget colors. The shift of attentional selectivity matched the shift of the memory representation of the target. Further, the bias in attentional capture was reduced when the nontarget colors were more distinct from the target. We discuss alternative accounts of the data, such as saliency-driven capture and the relational account of attentional capture (Becker, 2010), but conclude that optimal tuning theory provides the best explanation.