The present study completed a previous randomized trial that demonstrated the protective effect of 1-year psoas training on lumbar bone loss in postmenopausal women. Computerized tomography had been carried out at the beginning (CT1) and at the end (CT2) of this trial. In the present study, 67 women having completed the first trial were asked to practice psoas exercises (60 hip flexions in sitting position with a 5 kg weight on the knee) for 2 additional years with a third CT control at the end of this period (CT3). The aim of this complementary study was to assess the compliance rate and long-term effect on bone of daily psoas muscle training over a longer period. Twenty-one women performed this daily psoas training for 3 years from CT1 to CT3, and 14 acted as controls during the same period. Fourteen women were controls during the first year (from CT1 to CT2) but practiced psoas training during the following 2 years (from CT2 to CT3). Four women were psoas trained during the first year (from CT1 to CT2) and subsequently crossed over to the control group for the last 2 years. The compliance rate was 42%, with an attendance rate of 88%. The lumbar bone loss was lower in the 21 women trained over the 3 years (-3.26 +/- 28.45 mg/cm3) than in the 14 untrained women (-16.79 +/- 8.51 mg/cm3) (P = 0.02). The bone loss was not significantly reduced between the two periods of the study in the 12 women having been controls from CT1 to CT2 and having crossed over to the active training group from CT2 to CT3. Psoas training may be effective against lumbar bone loss. We conclude that specific training may play a contributing role in the preventive strategy to avoid osteoporosis.