We employed multilevel models to estimate the effects of neighborhood racial composition on birthweight, using a national urban sample of 1871 births to unmarried black mothers from the Fragile Families and Child Wellbeing birth cohort study. The mothers lived in 1181 census tracts with substantial variability in racial composition and poverty. Controlling for individual characteristics, census tract poverty, and city characteristics, the proportion black in the mother's census tract had no linear association with infant birthweight. There was an observed non-linear association, but the pattern was inconsistent with prevailing theories of how neighborhoods affect health. The results were robust to alternative measures, covariates, and sample restrictions and when accounting for the length of time the mother resided in her neighborhood.