The effect of anemia (hematocrit less than or equal to 0.34) on subsequent preterm birth was prospectively studied in 35,423 pregnancies. The incidence of preterm birth among women with and without anemia at each week during the third trimester was compared. Early in the third trimester, there was a weak association between anemia and preterm delivery. However, anemia early in the third trimester did not account for the substantial increase in preterm birth seen among black women. Anemia after 30 weeks' gestation was not associated with preterm birth. Among women delivering term infants weighing 2500 g or more, the mean hematocrit rose 0.029 among black women and 0.021 among white women from 25 weeks to term. Compared with hematocrits at 40 weeks' gestation, the odds ratios for anemia reached a maximum at 28 weeks and fell sharply as term approached. When the hematocrits of women in term labor were compared with those of women in preterm labor, a spurious dose-response effect for anemia was created. We conclude that anemia is not a strong factor in the pathogenesis of preterm birth and that comparison of hematocrits from women who are in preterm and term labor produces biased results.