Summary: We study observational learning in environments with congestion costs: an agent’s payoff from choosing an action decreases as more predecessors choose that action. Herds cannot occur if congestion on every action can get so large that an agent prefers a different action regardless of his beliefs about the state. To the extent that switching away from the more popular action reveals private information, it improves learning. The absence of herding does not guarantee complete (asymptotic) learning, however, as information cascades can occur through perpetual but uninformative switching between actions. We provide conditions on congestion costs that guarantee complete learning and conditions that guarantee bounded learning. Learning can be virtually complete even if each agent has only an infinitesimal effect on congestion costs. We apply our results to markets where congestion costs arise through responsive pricing and to queuing problems where agents dislike waiting for service.