Executive function (EF) improves between the ages of 3 and 5 and has been assessed reliably using the Dimensional Change Card Sort (DCCS), a task in which children first sort bivalent cards by one dimension (e.g., shape) and then are instructed to sort by a different dimension (e.g., color). Three-year-olds typically perseverate on the pre-switch dimension, whereas 5-year-olds switch flexibly. Labeling task stimuli can facilitate EF performance (Jacques & Zelazo, 2005; Kirkham, Cruess, & Diamond, 2003), but the nature of this effect is unclear. In 3 experiments we examined 2 hypotheses deriving from different theoretical perspectives: first, that labels facilitate performance in a more bottom-up fashion, by biasing attention to relevant task rules (Kirkham et al., 2003); and second, that labels aid performance in a more top-down fashion by prompting reflection and an understanding of the hierarchical nature of the task (Zelazo, 2004). Children performed better on the DCCS when labels referred to the relevant sorting dimension (Experiment 1). This was a function of the content of the labels rather than the change in auditory signal across phases (Experiment 2). Furthermore, labeling the opposite dimension only did not have a symmetrically negative effect on performance (Experiment 3). Together, these results suggest external, verbal labels bias children to attend to task-relevant information, likely through interaction with emerging top-down, endogenous control.