Adolescent health risk behaviors emerge in a broad peer context in which youngsters select and influence each other based on risk behaviors and other attributes. Advanced longitudinal social network analysis can disentangle such influence and selection processes in adolescent networks. Two important new issues will be examined studying theoretically postulated processes in relationship and behavior development in depth. First, previous research neglected effects of being embedded in complex multilevel social structures. Deindividuation theory for example states that in larger groups anonymity increases and personal responsibility diffuses, leading to reduced concern for social evaluation and loss of identity. After adapting existing software we will study network and behavior development using a multilevel oriented methodology going beyond strict nesting structures of traditional multilevel analysis. I will examine different levels such as the individual level, dyads, triads, members of personal networks, relationships between those members, and larger networks in which those members could be embedded. I hypothesize that adolescents in complex larger social structures are more susceptible to influence and that selection into complex larger social structures is more strongly based on behavioral similarities. Second, besides friendships, different relationship types may cause risk behaviors to spread. Solely focusing on friendship networks does not provide enough information to explain why adolescents take up risk behaviors and how these behaviors spread. It is important to focus on other linkages between adolescents like identification with similar youth cultures and being linked in online networks. Influence and selection processes based on risk behaviors are stronger when focusing on different types of network ties in addition to friendship ties. By studying existing and newly collected data, this project will increase our understanding of fundamental principles that drive social network formation among adolescents and increase our knowledge about the spread of risk behaviors in a multilevel context.