My thesis studies identification and estimation in network formation models. First, I study what can be learned from pairwise stable networks. Pairwise stability of a network gives strong identification power when I consider the probability that the observed network is pairwise stable. I propose a semiparametric maximum score estimator which is simple and computationally feasible. I apply the empirical model to social and economic networks in rural India, and find homophily patterns in village networks. Second, I propose a structural model of multigraph formation, where 1) individuals determine multiple types of links simultaneously; 2) all networks interact with each other; and 3) one or more networks are endogenous but not simultaneous. I extend the notion of pairwise stability to a multigraph, and show that the structural model is equivalent to a multinomial choice model. The presence of endogenous but not simultaneous networks is a source of an incomplete econometric model. Relying on partially identified econometric models, I characterize the sharp identification region of parameters by a finite set of moment inequalities. I apply the model to village networks and find that friendship affects risk sharing and favor exchange networks in the same direction. The last chapter studies an empirical model of network formation in the U.S. airline industry and investigates the size of network externalities. I assume that each airline builds a network that satisfies a weak notion of stability. That is, no airlines want to deviate from their current networks by a single route change. In this framework, I can use an entry game to investigate the airline industry and include network measures in the profit function to estimate network externalities. I find that when I control for the number of one-stop flights the effect of hub-size is larger than the case without considering one-stop flights.