Purpose: The existing literature on language outcomes in children born prematurely focuses almost exclusively on standardized test scores rather than discourse-level abilities. The authors of this study looked longitudinally at school-age language outcomes and potential moderating variables for a group of twins born prematurely versus a control group of twins born at full term, analyzing both standardized test results and language sample data from the population-based Western Reserve Reading Project (WRRP; Petrill, Deater-Deckard, Thompson, DeThorne, & Schatschneider, 2006). Method: Fifty-seven children born prematurely, at =32 weeks or <1,500 g, were compared with 57 children born at full term and were matched for age, gender, race, and parental education. Data included discourse-level language samples and standardized test results, collected at average ages 7, 8, and 10 years. The language samples were analyzed to yield a number of semantic and syntactic measures that were consolidated via factor analysis. Results: Regression models showed significant differences between the 2 groups for standardized test results, although the mean score for both groups fell in the normal range. For the discourse-level language measures, however, differences never reached statistical significance. Parental education was significantly associated with improved standardized test scores. Conclusions: These findings suggest that in the absence of frank neurological impairment, sophisticated semantic and syntactic skills may be relatively intact in the discourse-level language of children born prematurely. Implications for assessment, particularly the potential role of attention and executive function in standardized testing tasks, are reviewed.