This article estimates people's taste for living with own-ethnic-group neighbours using variation from a natural experiment in Singapore: ethnic housing quotas. I develop a location choice model that informs the use of policy variation from the quotas to address endogeneity issues well known in the social interactions literature. I assembled a dataset on neighbourhood-level ethnic proportions by matching more than 500,000 names in the phonebook to ethnicities. I find that all groups want to live with some own-ethnic-group neighbours but they also exhibit inverted U-shaped preferences so that once a neighbourhood has enough own ethnic neighbours, they would rather add a new neighbour from other groups. Welfare simulations show that about 30% of the neighbourhoods are within one standard deviation of the first-best allocation of ethnic groups.