Injured children (145 children between 2 and 13 years of age) were recruited from a hospital emergency room and were interviewed about the injury event soon afterward and then twice more at yearly intervals. Their transcripts were coded three ways: completeness of overall structural components of a prototypical injury event (who, when, where, etc.), amount of narrative detail (specifically, unique units of information), and the accuracy of both types of information. Completeness components were also categorized as central or peripheral, and narrative details were coded as pertaining to persons, objects, attributes, locations, or activities. Over time, children maintained consistent completeness scores; that is, the overall structure of the event stayed the same. However, they provided more elaborative detail of all types, and especially about attributes and activities. Only accuracy (of both types of information) deteriorated. Thus, different aspects of their interviews changed in different ways over two years. Implications for assessing changes over time in child witness reports are discussed.