The ability to recognize common objects from sparse information about geometric shape emerges during the same period in which children learn object names and object categories. Hummel and Biederman's (1992) theory of object recognition proposes that the geometric shapes of objects have two components--geometric volumes representing major object parts, and the spatial relations among those parts. In the present research, 18- to 30-month-old children's ability to use separate information about object part shapes and part relations to recognize both novel (Experiment 1) and common objects (Experiment 2) was examined. Children succeeded in matching novel objects on part shapes despite differences in part relations but did not match on part relations when there were differences in part shapes. Given known objects, children showed that they did represent the relational structure of those objects. The results support the proposal that children's representations of the geometric structures of objects are built over time and may require exposure to multiple instances of an object category. More broadly, the results suggest that the distinction between object part shape and part relations as two components of object shape similarity is psychologically real and developmentally significant. (Contains 3 figures and 1 table.)