This study compares three variations in how researchers construct middle childhood social networks: (1) with "friendships" or "affiliations" as a relational tie; (2) with children providing "self reports" of relationships, or in addition, "multi-informant reports" of relationships in which they are not involved; and (3) whether network computation is "correlational" or "distance-based." The sample was 357 fourth- and fifth-grade students in 17 classrooms. The strongest differences were between self-reported friendship and affiliative networks. Results showed that compared with affiliations, friendship networks had smaller groups, more isolates, and lower fall-to-spring stability. Agreement in social placement between friendship and affiliative networks was generally average, but poor for unpopular and aggressive children. Multi-informant affiliative networks were most robust in their positioning of aggressive children. Multi-informant centrality was uniquely uncorrelated with aggression. Network computation differences were not substantial. Discussion focuses on recommendations for research and the educational promise of network technology.