This paper studies the causal effect of English proficiency on residential location outcomes and the socioeconomic class of immigrants in England and Wales, exploiting a natural experiment. Based on the phenomenon that young children learn a new language more easily than older children, we construct an instrument for English proficiency using age at arrival in the United Kingdom. Taking advantage of a unique dataset, we measure the extent of residential segregation along different dimensions, and find that poor English skills lead immigrants to live in areas with a high concentration of people who speak their same native language, but not necessarily in areas with a high concentration of people of their same ethnicity or country of birth. This finding could suggest that, for immigrants with poor English proficiency, what matters for their residential location decision is language spoken by residents, as opposed to ethnicity or country of birth. We also find that language skills have an impact on the occupation-based socioeconomic class of immigrants: Poor English skills reduce the likelihood of being in the occupation-based class 'higher managerial and professional' and increase that of being in the class 'self-employment'.