A meta-analysis was done to measure the effect of physical activity on the bone mass of healthy postmenopausal women. All studies published between 1966 and 1996, in French or English, were reviewed for inclusion from Medline search, bibliographies of relevant studies, review articles and books. Studies had to be prospective intervention trials, randomized or not, evaluating the effectiveness of an exercise program of any duration, frequency and intensity, with a control group. Studies had to measure bone parameters and involve healthy postmenopausal women over 50 years of age who were free of symptomatic osteoporosis at the time of study entry. Effect sizes (ES) were calculated for each bone parameter and site measured in every eligible study according to Hedges and Olkin. DerSimonian and Laird's model was used to estimate overall effect sizes when combining studies. All analyses were bone parameter and site specific. Of 217 papers extracted from the literature, 187 did not meet eligibility criteria and 12 others were rejected. The two main reasons for rejection were that both genders were combined in the analyses and no exercise group without drug interaction was present. Eighteen studies were included for meta-analysis. Taking into account the frequency, duration, compliance rate and average age of the subjects, the programs were judged of moderate intensity and focused on walking, running, physical conditioning and aerobics. A significant effect of physical activity was detected on the bone mineral density at the L2-4 level of the lumbar column in studies published after 1991 (ES = 0.8745, p < 0.05). No effect could be seen, however, on forearm and femoral bone mass. Although applied to a small number of studies, this meta-analysis suggests that exercise programs in a population of postmenopausal women over 50 years of age are effective for preventing spinal bone mineral density loss at the L2-4 level. However, such programs do not have any effect on the forearm or femoral bone mass.