This study was designed to establish Bone Mineral Density (BMD) Reference Databases for multiple skeletal sites appropriate for the diagnosis and evaluation of osteoporosis (OP) in Chinese women. We recruited 2702 healthy Chinese women, 5–96 years of age, for BMD assessment. BMD values at multiple skeletal sites including anteroposterior (AP) and lateral (Lat) lumbar spine, hip, and forearm were measured by dual-energy X-ray absorptiometry (DXA) using a QDR 4500A device; results were analyzed according to age group using eight regression models. BMD Reference Databases (CWD) were established according to the best regression equation and compared with Hologic reference databases for “Oriental Women” (OWD). Results indicated that the cubic regression model was superior to the quadratic, linear, logarithmic, and exponential regression models, etc. for our purpose, with a determinate coefficient (R ²) of 0.363–0.650 (P = 0.000). We included 1636 female patients, aged 35–86 years, in our tests. In comparison with Hologic Reference Databases, the mean detection rate of OP in the newly established BMD Reference Databases for Chinese Women (CWD) was 16.0% ± 2.68% lower (range, 13.7%–20.5%) at the AP spine, 16.8% ± 11.0% lower (range, 3.5%–32.8%) at the Lat spine (except for L4), 18.7% ± 4.6% lower (range, 12.6%–24.2%) at the hip, and 14.3% ± 6.9% higher (range, 4.7%–24.3%) at the forearm. The difference in detection rates for OP was significant between the two reference databases (P = 0.000), which was consistent with the differences in peak BMD values and the biological variability between them. Based upon our data, we confirmed that the Hologic BMD Reference Databases for Oriental Women (OWD) were not suitable for the diagnosis of OP in Chinese women; the BMD Reference Databases for Chinese Women (CWD) established in this study would provide reliable diagnostic standards for detection of OP in the women of South China.