Observers find a target item more quickly when they have foreknowledge of target-defining attributes, such as identity, color, or location. However, it is less clear whether foreknowledge of attributes that define non-targets can also speed search. Munneke, Van der Stigchel, and Theeuwes (2008) cued observers to ignore a region of space where a target would not appear, and found these cues speeded responses. We replicated this effect, and found some evidence suggesting that speeded responses occurred because of inhibition of the distractor location rather than activation of the remaining possible target locations. Using a similar paradigm, we presented either a patch of color or printed verbal cues that instructed participants to ignore a particular color in a search task. We found that these cues resulted in less efficient search in both cases. Results from a probe-dot detection task revealed that this slowing occurred because, paradoxically, participants directed attention towards the item appearing in the to-be-ignored color. We found that participants could effectively ignore a non-target color when items matching that color were previewed for at least 800 ms; thus inhibition can occur following selection. Ignore cues slowed search when the items were in motion, or when the to-be-ignored feature composed a background segment of the display rather than being bound to a display item. However, ignore cues were less detrimental (and in a few cases slightly helpful) when the to-be-ignored item was a salient singleton that automatically captured attention. We propose a "Search and Destroy" model to account for this behavior: when an observer attempts explicit feature-based inhibition, they seek out items matching the to-be-ignored feature in order to subsequently inhibit those items.