Attention at encoding plays a critical and ubiquitous role in explicit memory performance, but its role in implicit memory performance (i.e., priming) is more variable: some, but not all, priming effects are reduced by division of attention at encoding. A wealth of empirical and theoretical work has aimed to define the critical features of priming effects that do or do not require attention at encoding. This work, however, has focused exclusively on priming effects that are beneficial in nature (wherein performance is enhanced by prior exposure to task stimuli), and has overlooked priming effects that are costly in nature (wherein performance is harmed by prior exposure to task stimuli). The present study takes up this question by examining the effect of divided attention on priming-induced costs and benefits in a speeded picture-naming task. Experiment 1 shows that the costs, but not the benefits, are eliminated by division of attention at encoding. Experiment 2 shows that the costs (as well as the benefits) in this task are intact in amnesic participants, demonstrating that the elimination of the cost in the divided attention condition in Experiment 1 was not an artifact of the reduced availability of explicit memory in that condition. We suggest that the differential role of attention in priming-induced performance costs and benefits is linked to differences in response competition associated with these effects. This interpretation situates the present findings within a theoretical framework that has been applied to a broad range of facilitatory priming effects.