The major effect of weightbearing exercise on adult bone mass may be exerted during childhood. We examined the relationship between reported hours of ballet classes per week undertaken as a child and adult bone mineral density (BMD) at the hip, spine, and forearm. We performed a retrospective cohort study in 99 female retired dancers (mean age 51 years, SD 14 years) and 99 normal controls, derived from a twin study, matched hierarchically for age, height, weight and menopausal status. Starting age of ballet was recalled and weekly hours of ballet as a child was self-reported on two occasions. BMD was measured using dualenergy X-ray absorptiometry and reported as a Z-score. Self-reported hours of ballet class undertaken per week at each age between 10 and 12 years was positively associated with a difference in BMD between dancers and controls at both the femoral neck site (beta = 0.73, p = 0.001) and the total hip site (beta = 0.55, p < 0.01). These associations were unaffected by adjustment for covariates including measures of adult activity (current physical activity, years of fulltime ballet), measures of menstrual disturbance (age of menarche, history of irregular menses), dietary history (calcium intake as a child, adolescent or adult) or lifestyle factors (lifetime smoking, lifetime alcohol). Although starting age of ballet was negatively associated with weight-adjusted within-pair hip BMD difference, it was no longer associated after adjustment for weekly hours of ballet. There was no relationship between hours of ballet undertaken as a child and differences in BMD at the lumbar spine or upper limb, at any age. Our data suggest that classical ballet classes undertaken between the ages of 10 and 12 years are independently and positively associated with a difference in hip BMD between dancers and controls. The findings are consistent with the hypothesis that this age range identifies a stage of development when the proximal femur is particularly responsive to weightbearing exercise.