We examine the association between late-stage breast cancer diagnosis and residential poverty in Detroit, Atlanta, and San Francisco in 1990 and 2000. We tested whether residence in census tracts with increasing levels of poverty were associated with increased odds of a late-stage diagnosis in 1990 and 2000 and found that it was. To test this, we linked breast cancer cases from the Surveillance, Epidemiology, and End Results cancer registries with poverty data from the census. Tracts were grouped into low, moderate, and high poverty based on the percentage of households reporting income below the poverty level. While late-stage breast cancer rates and the number of women living in high and moderate-poverty areas declined absolutely between 1990 and 2000, estimates from our combined three-city model showed that odds of a late-stage diagnosis remained stubbornly elevated in increasingly poor areas in both years. Non-Hispanic black women faced higher odds of a late-stage diagnosis relative to non-Hispanic white women in both years. In separate regressions for each city, the odds ratios affirm that combining data across cities may be misleading. In 1990 and 2000, only women living in moderately poor neighborhoods of San Francisco faced elevated odds, while in Detroit women in both moderate- and high-poverty areas faced increased likelihood of late-stage diagnosis. In Atlanta, none of the poverty measures were significant in 1990 or 2000. In our test of physician supply on stage, an increase in the number of neighborhood primary care doctor's offices was associated with decreased odds of a late-stage diagnosis only for Detroit residents and for non-Hispanic whites in the three-city model.