The way that children construct the representation they use to solve transitive inference problems was examined. Forty-eight children 4.5 to 5 years old and 48 children 6 to 7 years old were asked to learn either a three-item series or a four-item nonseries. They were asked to learn the relationships between different colors of faces that were all the same size; on each trial, they were asked which face was bigger or which face was smaller. During the testing phase, two visual feedback conditions were included to provide a test of whether children use absolute size information to answer the inference question. In the linguistic feedback condition, the child was told which face was bigger or smaller. It was found that the linguistic condition was harder than the visual absolute condition. Overall, the findings suggest that children do not remember absolute size information. Older children appeared more successful than younger children in learning the premise pairs in the series condition. Younger children were more successful in learning the premise pairs in the nonseries condition than in the series condition. The nonseries was easily learned by both groups since it did not have a common middle term and did not require seriation skills. The main conclusion is that there are developmental changes in children's abilities to order a series. It appears that children can use trial-and-error processes to learn a series and that they can use a linear order to make inferences. Implications of the findings to views proposed by Piaget and by Trabasso are also addressed. (SW)