We investigate the role of information frictions in migration. Using novel moment inequalities and data on internal migration in Brazil, we estimate worker preferences and migration costs while allowing for unobserved worker-specific information sets. We find that common estimation procedures overestimate migration costs and underestimate the importance of expected wages in migration decisions. Model specification tests indicate that workers often have limited information on location-specific wages. However, those living in regions with better internet access and larger populations have more precise wage information, and information precision decreases with distance. According to our estimated model, workers' limited wage information plays a quantitatively important role in reducing migration flows and worker welfare, and limits the effect of policies that reduce migration costs.