Birth characteristics predict a range of major physical and mental disorders, but findings regarding eating disorders are inconsistent and inconclusive. This total-population Swedish cohort study identified 2,015,862 individuals born in 1975-1998 and followed them for anorexia nervosa, bulimia nervosa, and eating disorder not otherwise specified until the end of 2010. We examined associations with multiple family and birth characteristics and conducted within-family analyses to test for maternal-level confounding. In total, 1,019 males and 15,395 females received an eating disorder diagnosis. Anorexia nervosa was independently predicted by multiple birth (adjusted hazard ratio = 1.33, 95% confidence interval: 1.15, 1.53) for twins or triplets vs. singletons) and lower gestational age (adjusted hazard ratio = 0.96, 95% confidence interval: 0.95, 0.98) per extra week of gestation, with a clear dose-response pattern. Within-family analyses provided no evidence of residual maternal-level confounding. Higher birth weight for gestational age showed a strong, positive dose-response association with bulimia nervosa (adjusted hazard ratio = 1.15, 95% confidence interval: 1.09, 1.22, per each standard-deviation increase), again with no evidence of residual maternal-level confounding. We conclude that some perinatal characteristics may play causal, disease-specific roles in the development of eating disorders, including via perinatal variation within the normal range. Further research into the underlying mechanisms is warranted. Finally, several large population-based studies of anorexia nervosa have been conducted in twins; it is possible that these studies considerably overestimate prevalence.