Anorexia nervosa occurs early in life and predisposes to osteoporosis. Exercise may be protective. We asked: (1) Does failure to attain peak bone density contribute to the deficit in bone density? (2) Does oral contraceptive use protect against osteoporosis? (3) Is any protective effect of exercise confined to weight-bearing sites? Areal bone density (g/cm2) and body composition were measured by dual x-ray absorptiometry in 65 patients with anorexia nervosa and 52 controls. Comparing the 12 patients with primary amenorrhea and the 37 patients with secondary amenorrhea, bone density (mean ± SEM) at the lumbar spine was 0.88 ± 0.04 versus 1.06 ± 0.03 (P = 0.001), respectively. Bone density at the femoral neck was 0.80 ± 0.04 versus 0.92 ± 0.03 (P < 0.05), respectively. These values differed before, but not after, adjusting for the respective duration of illness (73.0 ± 10.3 versus 34.1 ± 4.8 months, P < 0.001) and fat-free mass (31.6 ± 1.3 versus 35.4 ± 0.5 kg, P < 0.01). Bone density at the lumbar spine in the 16 patients with 31.8 ± 8.3 months of contraceptive exposure was higher than in the 49 patients with no contraceptive exposure (1.14 ± 0.05 versus 1.02 ± 0.02 P < 0.02) but was lower than in controls (1.14 ± 0.05 versus 1.27 ± 1.02, P < 0.01). No protective effect of contraceptive exposure was detectable at the femoral neck. Nevertheless, bone density at both sites correlated positively with the duration of contraceptive exposure (lumbar spine, R 0.63, P < 0.02; femoral neck, R = 0.66 P < 0.01) and negatively with duration of amenorrhea (lumbar spine, R = −0.53, P < 0.01; femoral neck, R = −0.49, P < 0.01, respectively). In the 19 patients who exercised vigorously, bone density was higher than in the 30 sedentary patients at the femoral neck (0.97 ± 0.04 versus 0.85 ± 0.03, P < 0.05), but not at the lumbar spine. Anorexia nervosa of earlier onset is likely to impose a greater fracture risk because of larger deficits in bone density. The deficits are more likely to be due to bone loss than low peak bone density, although a contribution to the deficit by low peak bone density cannot be excluded. Bone density of the lumbar spine may be more sensitive to estrogens than bone density of the proximal femur, and the proximal femur may be more sensitive to weight-bearing exercise. Because partial protection against osteoporosis may be obtained at the spine using oral contraceptives and at the proximal femur by weight bearing, these modalities may warrant consideration as part of the treatment of this illness.