We study a coordination game among agents in a network. The agents choose whether to take action (e.g., adopting a new technology) in an uncertain environment that yields increasing value in the actions of neighbors. We develop an algorithm that fully partitions the network into communities (coordination sets) within which agents have the same propensity to adopt. Our main finding is that a novel measure of network connectedness, which we term “social connectedness”, determines the propensity to adopt for each agent. Social connectedness captures both the number of links each agent has within her community (interconnectedness) as well as the number of links she has with members of other communities who have a higher propensity to adopt (embeddedness). There is a single coordination set if and only if the network is balanced —that is, the average degree of each subnetwork is no larger than the average degree of the network. Finally, we demonstrate that contagion is localized within coordination sets, such that a shock to an agent uniformly affects this agent and all members of her coordination set but has no impact on the other agents in the network.