Seventy years ago Paul Lazarsfeld and colleagues found empirical evidence for a significant influence of the social context on media effects by interpersonal communication. Subsequently various theories in mass media science incorporated the social context as an independent or dependent variable. However, there is little empirical research that addresses the dynamics of media use within a social context using network analytic methods. This study considers the social context as both an independent and dependent variable in a dynamic network process in order to disentangle social selection and influence processes. Hence, on one hand it tests whether the intensity of TV use and the use of specific TV genres predict the selection of conversation partners in social groups (social selection process). On the other hand it tests if individuals’ social context predicts their TV use (social influence process). Here, social context is defined as the conversation structure as well as the media use of all other persons in the social group. The research design includes a four wave panel survey on interpersonal communication networks and TV use of 707 students (age 13–16) in 29 Swiss school classes. The stochastic actor-based models tested with the program SIENA support the hypothesized selection processes. Conversation partners about TV programs are selected according to the similarity of their TV use. In contrast to this finding, the widely held assumption that individuals are influenced by their social network is not supported. Some inconclusive evidences suggest a possible social influence process on the level of TV intensity but not on the level of specific TV genres. Network-autocorrelation of conversation ties and TV use has therefore primarily be accredited to social selection processes and not social influence. Furthermore, the results show that avid viewers tend to talk more often about TV programs (ego effect), that avid viewers are more likely to be addressed (alter effect), and a general tendency to talk to persons with the same program preference (similarity effect). This challenges the classic idea of a two-step flow of communication in which intensive media users which are well informed would provide information to occasional users which are less conversant with a topic.