Working memory (WM) is enhanced for items that capture attention, but little is known about how it achieves this effect. In this talk, I will provide behavioral and neural evidence that bottom-up attention is helpful only when perceptual selection demands are high. When multiple information sources compete, bottom-up attention prioritizes the location at which encoding should begin. When encoding order is set, bottom-up attention has little or no benefit to working memory. In contrast, voluntary attention improves working memory regardless of encoding order. Neural evidence from fMRI and ERP supports this hypothesis. These results suggest that bottom-up attention improves the probability that items enter WM, but voluntary attention is a better predictor of the quality of the representation.