This paper addresses two questions: 1) What is the relation of hemoglobin in the second gestational month to preterm birth and low birth weight? 2) How does the relation differ when hemoglobin in the fifth or eighth month or the lowest pregnancy hemoglobin are examined in place of first trimester values? These relations were examined prospectively in 829 women from Shanghai, China in 1991–1992. The population was nearly homogeneous by race, parity, antenatal care, and smoking. Rates of birth outcomes were compared between hemoglobin categories based on 10 g/liter groupings, with 110–119 g/liter as the reference group. Rates of low birth weight and preterm birth (but not small-for-gestational age) were related to early pregnancy hemoglobin concentration in a U-shaped manner. The relative risks (95% confidence intervals) for preterm birth in women by g/liter of hemoglobin were 2.52 (0.95–6.64) for ≥130 g/liter, 1.11 (0.41–2.99) for 120–129 g/liter, 1.64 (0.77–3.47) for 100–109 g/liter, 2.63 (1.17–5.90) for 90–99 g/liter, and 3.73 (1.36–10.23) for 60–89 g/liter. Use of hemoglobin values in the fifth or eighth month attenuated the association with preterm birth. When lowest pregnancy hemoglobin values were used, the association of anemia with both outcomes was obscured, and risk of preterm birth at high hemoglobin values increased dramatically. Am J Epidemiol 1998; 148: 998–1006.