When searching for a "pop-out" target, interference from a salient but irrelevant distractor can be reduced or even prevented under certain circumstances. Here, five experiments were conducted to further our understanding of three different aspects of top-down interference reduction: first, whether or not qualitatively different search modes can account for different reduction patterns; second, whether distractor practice plays a causal role in reduction; and third, how specific reduction is, that is, whether interference by intradimensional distractors can be reduced as effectively as interference by cross-dimensional distractors. The results provide evidence that interference reduction does not critically depend on the implementation of a feature search mode, but rather on practice with the distractor, that is, the acquisition of an effective suppression strategy. In addition, they suggest that interference reduction is based on hierarchically organized feature weighting ("dimension weighting"), rather than on completely independent feature weighting. (Contains 4 tables, 7 figures and 4 footnotes.)