In this dissertation I analyze the economics of racial segregation. Each chapter provides theoretical models and empirical methods to analyze the separation of racial groups in several contexts. Chapter 2 propose a method for measuring residential segregation using techniques from the spatial statistics literature. The available indices of segregation depend on a partition of the city in neighborhoods: given the spatial distribution of racial groups, different partitions translate in different levels of measured segregation. I propose a location-specific index, that maps individual coordinates to local level of segregation. The segregation of the metropolitan area is measure as the average individual segregation. Therefore, the level of segregation measured according to my approach is independent from arbitrary partitions. I show that this method provides a different ranking of cities' segregation than the traditional neighborhood-based measures. The method estimates the entire distribution of segregation across individuals and I provide evidence that high levels of aggregate segregation are the consequence of very few highly segregated neighborhoods. Using the spatial indices, I show evidence of the negative effect of segregation on individual outcomes of minorities. Chapter 3 and 4 analyze segregation in social networks. In Chapter 3, I develop and estimate a structural model of strategic network formation with heterogeneous agents. Structural estimation of strategic models of network formation is challenging, since these models usually have multiple equilibria. I present a dynamic model where the network is formed sequentially: each period an individual has the opportunity to update his linking strategy. This generates a sequence of networks that converges to a unique stationary equilibrium. I characterize the equilibrium as providing the likelihood of observing a specific network structure in the long run. However the estimation is complicated, since the likelihood is proportional to a normalizing constant that cannot be evaluated or approximated with precision. To overcome this problem, I propose a Bayesian Markov Chain Monte Carlo method that allows estimation of the posterior without evaluating the likelihood. I study segregation in social networks using data from Add Health, a survey of US high schools, containing detailed information on school friendship networks. I find that students prefer interactions with individuals of the same race. The simulation of several busing programs shows that perfect integration across schools may not be optimal. An equalization of racial shares across schools may increase segregation and decrease welfare. In Chapter 4, I focus on an alternative estimation method. I propose an approximate Maximum likelihood estimation strategy. Assuming the utilities are linear in parameters, it can be shown that the Maximum likelihood maximization problem has the same solution of a system of nonlinear equations, which I solve using a stochastic approximation algorithm. To perform the stochastic approximation, I develop an algorithm to generate samples from the stationary equilibrium of the model. The algorithm is a variant of the Simulated tempering and allows fast convergence to the equilibrium distribution, decreasing the computational costs of estimation. Using Add Health data, I confirm the results of Chapter 3.