Social networks are known to influence migration decisions, but connections between individuals remain usually unobserved. Surveys conducted by Gallup in 147 countries provide information on migration intentions and on distance-one connections in each destination. The distribution of distance-one connections mirrors the one of migrant stocks, and intentions are informative about actual decisions. The estimation of originspecific conditional logit models reveals that distance-one connections can alter the ranking of most pairs of destinations. We test the validity of the distributional assumptions that underlie identification and perform extensive robustness checks, thus mitigating the concerns about the threats to identification posed by unobservables.