Migrant networks play an important role in explaining the size and structure of migration flows. They affect the private costs and benefits of migration (assimilation channel) and lower legal entry barriers through family reunification programs (policy channel). This paper presents a micro-founded identification strategy allowing to disentangle the relative importance of these two channels. Our empirical analysis exploits US immigration data by metropolitan area and country of origin. We first find that the elasticity of migration flows to network size is around one. More interestingly, we show that the policy channel accounted at most for a quarter of this elasticity in the 1990s, and the magnitudes of the total network effect and the policy channel are greater for low-skilled migrants. Our results are strongly robust to sample selection, identification assumptions, and treatment for unobserved bilateral heterogeneity. Furthermore, the policy channel was stronger in the 1990s than in the 1980s, possibly reflecting the changes in the US family reunification policies. We conclude that the government capacity to curb the migration multiplier exists, but is limited.