A mathematical model with adequate clinical accuracy would be useful for predicting life-long bone mass behaviour in various skeletal sites. We have tested the performance of a multivariate nonlinear model, recently developed by an Italian research group for predicting lumbar bone mass and density, in a sample of 239 pre- or post-menopausal healthy Finnish women whose lumbar bone mineral density (BMD) was determined by dual-energy X-ray absorptiometry. The model, which was based on height, weight, body mass index, age and years since menopause, predicted poorly (R-squared = 0.14) the measured BMD. Furthermore, it compressed strongly the actual variation of the BMD values resulting in a considerable overestimation or underestimation (about 30%) of low or high BMD values, respectively. Our results indicated that the predictive value of a simple model is not clinically acceptable and therefore the usefulness of this kind of model remains questionable.