Changing the way in which children respond significantly improves performance across a variety of cognitive domains. However, the basis for this response-mode effect is unknown. To address this issue, the current study tested 78 preschool children on a measure of executive functioning (the windows task: Hala & Russell, 2001). The task requires children to point to an empty box to receive a box with a treat in it. There were four versions of the task, differing only in the way in which children made their responses. Children in the baseline condition, who pointed with their finger, performed poorly. However, children in three alternative response-mode conditions won the treat significantly more often. Strikingly, even children who pretended to be pointing with an arrow – but were in reality pointing with their finger – performed significantly better than baseline. We suggest that non-standard response modes encourage children to reflect on their actions and that this reduces the number of unreflective errors they make. The findings demonstrate that response-mode effects do not depend on the presence of a physical means of responding (such as an arrow), but can be achieved via a purely cognitive means (such as engaging in pretence).