<p>This paper identifies and estimates the effects of student-level social spillovers on standardized test performance in New York City (NYC) elementary schools. We leverage student demographic data to construct within-classroom social networks based on shared student characteristics, such as gender or ethnicity. Rather than aggregate shared characteristics into a single network matrix, we specify additively separate network matrices for each shared characteristic and estimate city-wide peer effects for each one. Our work is based on the common assumption that more shared characteristics between peers imply stronger network connections. We test this hypothesis using a collection of common and novel characteristics, answering which shared characteristics exhibit the strongest peer effects and by how much. Conditional on being in the same classroom, we find that the most influential networks are shared gender and primary language spoken at home. We show that altering classroom composition changes the impact of these networks. Particularly, low linguistic diversity is correlated with low impact for shared language. As the influence of the shared language network declines, the gender network becomes more influential, indicating that these networks are substitutes. We discuss identification of the model and its implications for within- and between-group test performance gaps along several demographic traits.</p>