The impacts of social networks on changes in contraception in rural Kenya are investigated using special data from a longitudinal household survey. An analytic model, informed by detailed knowledge of the setting, yielded estimates that indicate that (1) social networks have substantial effects even after unobserved factors (e.g., homophily) that may determine social networks are controlled; (2) controlling for these unobserved factors may substantially alter the estimated effects of networks (these controls were not used in previous studies); (3) network effects are important for both men and women; and (4) network effects are nonlinear and asymmetric, suggesting that networks provide information primarily through social learning, rather than by exerting social influence.