The issue of whether salient distractors capture attention has been contentious for over 20 years. According to the salience-driven selection theory, the most salient location in the display is detected preattentively, after which attention is deployed automatically to that location. By other accounts, attentional deployment to the location of an item is contingent upon the task-relevance of that item. In the present work, six experiments employed the event-related potential (ERP) technique to examine the salience-driven selection and other theories of visual search. The experiments adopted additional singleton search, pop-out detection, and attentional-window paradigms. The ERP evidence obtained from the additional-singleton paradigm indicated that although the location of a salient item – whether a target or a distractor – was registered relatively early, the salient distractor did not capture attention consistently. Moreover, when the features of the salient distractor were held constant, observers were occasionally able to suppress the location of the distractor, thereby improving the efficiency of the search. The ERP evidence obtained from a Go/No-Go pop-out detection task indicated that attention was deployed to the location of a pop-out item only when a decision to search was made and, thus, that item was relevant to the observer’s goals. The ERP evidence obtained from the attentional-window paradigm indicated that goal-driven control over stimulus salience could extend to the items located within the observer’s attentional window. The present results suggest that while the locations of a limited number of salient items in the display can be registered on an early salience map, there is some goal-driven control over attentional deployment to the location of salient items or suppression of such locations. Factors that are potentially important in this dynamic control include the task-relevance of the search display, the predictability of distractor features, and inter-trial changes in target and distractor features and their task-relevance.