I study the impact of echo chambers on social learning outcomes in a setting featuring agents who are aware of the presence of echo chambers and are fully Bayesian. I introduce a novel sequential learning model, where the agent’s private signal is correlated with her neighbors’ signals conditional on the state of the world (echo chambers), and agents sequentially learn about the changing state of the world by observing their neighbors’ actions and their own private signals. I introduce a measure of echo chambers in this social learning environment, that depends both on the network homophily and the signal correlation, and analyze how the presence of echo chambers affects opinion polarization and asymptotic learning accuracy.