We estimate the chances of poor and non-poor children getting places in good schools, analysing the relationship between poverty, location and school assignment. Our dataset allows us to measure location and distance very precisely. The simple unconditional difference in probabilities of attending a good school is substantial. We run an analysis that controls completely for location, exploiting within-street variation and controlling for other personal characteristics. Children from poor families are significantly less likely to go to good schools. We show that the lower chance of poor children attending a good school is essentially unaffected by the degree of choice. (Contains 2 figures and 7 tables.)