The paper investigates the impact of ethnic segregation on the life chances of low-income African American and Latino children, focusing on whether it is the ethnic composition of the neighbourhood per se that matters or other, correlated aspects of the residential environment. The approach links the consequences of segregation and neighbourhood effects literatures by arguing that metropolitan segregation forces directly shape children’s intra-neighbourhood ethnic exposure and indirectly shape their exposure to non-ethnic aspects of neighbourhood. Associations between a wide range of neighbourhood characteristics and children’s health, exposure to and engaging in violence, educational and fertility outcomes are quantified using a natural experiment, thereby permitting valid causal inferences. Data analysed come from a retrospective survey of Denver (CO) Housing Authority (DHA) residents. The analysis avoids parental geographic selection bias because DHA’s assignment of households to neighbourhoods mimics a random process. Logit models stratified by ethnicity show that growing up amid concentrations of African American residents is associated with a variety of adverse outcomes for low-income Latino and (especially) African American children, though outcomes associated with concentrations of Latino residents are more mixed. Virtually all of the negative associations disappear, however, when other aspects of the residential context are controlled, and several positive ones persist. The adverse developmental consequences of ethnic segregation appear to be generated primarily in Denver by concentrating minority children in neighbourhoods with higher rates of property crime and lower occupational prestige.