Children's solutions to spatial problems pose an intriguing puzzle in development. For some time, difficulties with such tasks have been thought to index a pervasive characteristic of children's thought, called “egocentrism,” but recent research has shown clearly that at least by 2 years of age, children are not egocentric in the sense of not knowing that other people see displays differently. This chapter provides the argument that perspective-taking tasks and other spatial problems, such as mental rotation, are difficult for reasons that include the demands the particular problem makes on the usual form of representation of space. From at least 5 years of age, and perhaps earlier, this representation seems to be one in which small movable targets are encoded in relation to a framework of fixed landmarks, rather than in relation to each other. Such coding makes ecological sense, because movable items, almost by definition, make poor reference points for the location of other objects. Thus, spatial coding does not seem to change across middle childhood in the fashion discussed by Piaget. What remains for the future is to explore whether frameworks of landmarks are used soon after the beginning of free locomotion, and how location coding changes, if it does, in the first few years of life.