In visual search, a set of distractor items can be suppressed from future selection if they are presented (previewed) before a second set of search items arrive. This visual marking mechanism provides a top-down way of prioritizing the selection of new stimuli, at the expense of old stimuli already in the field (Watson & Humphreys, 1997). Typically, this preview benefit has been examined by measuring the efficiency of detecting a target item contained within a newly presented set. Such work has led to the suggestion that the capacity of the visual marking mechanism is extremely high. Here we present five experiments which measured performance for selecting and responding to all the new stimuli rather than a single target item within the new set. The findings illustrate that when selecting and responding to all new items intentionally trying to prioritize new stimuli has a capacity limit of approximately six to seven items and that this limit depends partly on properties of the stimuli. The findings are discussed in terms of mechanisms of time-based selection, attentional capacity limits, and the task demands of multiple item selection.