Future-oriented thought is ubiquitous in humans but challenging to study in children. Adults not only think about the future but can also represent a future state of the world that differs from the present. However, behavioral tasks to assess the development of future thought have not traditionally required children to do so as most can be solved based solely on representations of the present. To overcome this limitation, we modified an existing task such that children could not simply rely on a representation of the present to succeed (i.e., the correct answer for "right now" was different than the correct answer for "tomorrow"). A sample of 117 4- to 7-year-olds (64 girls and 53 boys) from Ottawa, Canada, and surrounding area, who were predominantly European Canadian (78.6% of sample) and had a family income of over $100,000 CAN (66.1% of sample) participated. Children remembered the information required to solve our task, and there were age-related changes in performance, but only 7-year-olds made an adaptive future-oriented decision significantly more often than chance. With the task modification removed (so the correct answer for the present and the future was the same), even 4-year-olds were above chance. Our work challenges the notion that starting at age 4, children solve behavioral tasks of future thinking by acting on their representations of the future.