The adipocyte-derived hormone adiponectin promotes insulin sensitivity and anti-atherogenic effects. Serum adiponectin levels are low in obese, insulin-resistant individuals but high in lean states such as caloric restriction (CR) or anorexia nervosa. Indeed, despite being produced exclusively by adipose tissue, serum adiponectin inversely correlates with % body fat. The basis for this so-called ‘adiponectin paradox’ is yet to be resolved. The discovery of adiponectin preceded our current understanding of white adipose tissue (WAT) as a major endocrine organ. In contrast, research has largely neglected another adipose depot: bone marrow adipose tissue (MAT). In humans MAT accounts for ≤ 70% of marrow volume and, in contrast to WAT, MAT increases in states of leanness such as CR or anorexia nervosa. Moreover, pharmacological agents such as thiazolidinediones or fibroblast growth factor-21 increase marrow adiposity. Notably, each of these conditions is also associated with increased serum adiponectin. Based on these observations, we investigated the hypothesis that MAT is a major source of serum adiponectin. To do so, we first isolated WAT and MAT from rabbits, mice and humans. Rabbits were sedated by i.m. injection of ketamine (40 mg/kg) and xylazine (5 mg/kg) before euthanising by i.v. injection of pentobarbital (65 mg/kg); mice were euthanised by CO2 asphyxiation followed by cervical dislocation; and human tissues were obtained with consent from patients undergoing lower-limb amputation. Immunoblotting of tissue lysates revealed that, in rabbits (n = 9) and mice (n = 6), adiponectin expression is markedly higher in MAT than in WAT. Although this was not consistently observed in human tissues (n = 7), human MAT had increased expression of Ero1-Lα, a protein chaperone that promotes adiponectin secretion. Indeed, adiponectin secretion from cultured MAT explants was far higher than that from WAT explants of rabbits (n = 9) or humans (n = 4), as assessed by immunoblotting. We next investigated if blocking MAT formation affects serum adiponectin levels. To do so we used Ocn-Wnt10b mice, in which the anti-adipogenic effector Wnt10b is transgenically expressed in bone. Expression of adiponectin mRNA in WAT did not differ between Ocn-Wnt10b (1.00 ± 0.21; n = 10) and control mice (0.91 ± 0.11; n = 10); however, Ocn-Wnt10b mice resisted increases in MAT and serum adiponectin with CR. Moreover, we found that, in mice (n = 24), total serum adiponectin negatively correlates with WAT mass (R2 = 0.441; P = 0.031) but positively correlates with MAT volume (R2 = 0.524; P = 0.009). In summary, our observations support the hypothesis that MAT expansion is required for increased serum adiponectin with CR, suggesting that MAT may be a major source of serum adiponectin. Thus, adiponectin production from MAT may account, at least in part, for the adiponectin paradox.