Many metropolitan areas in the United States display substantial racial segregation and substantial variation in incomes and house prices across neighborhoods. To what extent can this variation be summarized by a small number of representative (or synthetic) neighborhoods? To answer this question, U.S. neighborhoods are classified according to their characteristics in the year 2000 using a clustering algorithm. The author finds that such classification can account for 37 percent of the variation with two representative neighborhoods and for up to 52 percent with three representative neighborhoods. Furthermore, neighborhoods classified as similar to the same representative neighborhood tend to be geographically close to each other, forming large areas of fairly homogeneous characteristics. Representative neighborhoods seem a promising empirical benchmark for quantitative theories involving neighborhood formation.