BACKGROUND: Racial residential segregation in the United States has been linked to racial differences in birth outcomes, with studies reporting associations between segregation and birth weight. However, this relationship is likely confounded, and many individual and neighborhood-level covariates included in previous models are likely mediators, potentially obscuring any causal impact of segregation on birth weight. METHODS: We compiled a record of non-Hispanic black and white singleton births to US-born/resident mothers in 2000, linked to segregation indices at the metropolitan statistical area (MSA) level in the non-Southern US. Segregation was measured via the dissimilarity index. The outcomes were individual-level birth weight and the metropolitan statistical area-level black/white gap in birth weight. We instrumented for segregation using the railroad division index. We compared race-stratified ordinary least squares models to two-stage least squares models, with cluster robust standard errors. RESULTS: We estimated a 1.2 g decrease in black birth weight for every one-percentage point increase in segregation (95% confidence interval [CI]: -1.9, -0.50) via ordinary least squares but a 2.8 g decrease (95% CI: -6.0, 0.48) using two-stage least squares. For white infants, our ordinary least squares estimate was 0.53 (95% CI: -0.23, 1.3), and our two-stage least squares estimate was in the opposite direction (-0.68, 95% CI: -3.5, 2.1). CONCLUSIONS: Ordinary least squares estimates may understate the effect of segregation on birth weight in blacks. Evidence from instrumental variable models was consistent with a causal impact of segregation on black birth outcomes, but estimates were imprecise and may be affected by weak instrument bias.