This study presents new estimates of the impact of social networks on attitudes and behavior in two areas, family-planning and AIDS. The study explicitly allows for the possibility that social networks are not chosen randomly, but rather that important characteristics such as unobserved preferences and community characteristics determine not only the outcomes of interest but also the conversational networks in which they are discussed. To examine this issue, longitudinal survey data from rural Kenya are used. The major findings are: First, the endogeneity of social networks can substantially distort the usual cross-sectional estimates of network influences. Second, the estimates indicate that social networks have significant and substantial effects even after controlling for unobserved factors that may determine the nature of the social networks. Third, these network effects generally are nonlinear and asymmetric. In particular, they are relatively large for individuals who have at least one network partner who is perceived to be using contraceptives or to be at high risk of HIV/AIDS.