The paper investigates social-learning when the information structure is not commonly known. Individuals repeatedly interact in social-learning settings with distinct information structures. In each round of interaction, they use their experience gained in past rounds to draw inferences from their predecessors’ current decisions. Such adaptation yields rational behavior in the long-run if and only if individuals distinguish social-learning settings and receive rich feedback after each round. Limited feedback may lead individuals to imitate uninformed predecessors. Moreover, adaptation across social-learning settings renders Bayes’ rule payoff-inferior compared to non-Bayesian belief updating rules and suggests that belief-updating rules are heterogeneous in the population.