Studies of the effect of exercise programs on bone mass appear inconsistent. Our objective was to systematically review and meta-analyze randomized trials of the effect of exercise on bone mass in pre- and postmenopausal women. A computerized MEDLINE search was conducted for the years 1966-1997. Thirty-five randomized trials were identified. Meta-analytic methods were used to statistically pool results of studies of the effect of impact (e.g., aerobics) and non-impact (e.g., weight training) exercise on the lumbar spine and femoral neck. The most studied bone site was the lumbar spine in postmenopausal women (15 studies), where both impact [1.6% bone loss prevented, 95% confidence intervals (CI): 1.0%-2.2%] and non-impact (1.0%, 95% CI: 0.4%-1.6%) exercise programs had a positive effect. Results for the lumbar spine in premenopausal women (eight studies) were similar: 1.5% (95% CI: 0.6%-2.4%) less bone loss (or net gain) after impact exercise and 1.2% (95% CI: 0.7%-1.7%) after non-impact exercise. Impact exercise programs appeared to have a positive effect at the femoral neck in postmenopausal women (five studies), 1.0% (95% CI: 0.4%-1.6%) bone loss prevented, and possibly in premenopausal women, 0.9% (95% CI: -0.2%-2.0%) bone loss prevented. There were too few trials to draw conclusions from meta-analyses of the effect of non-impact exercise on the neck of femur. This systematic review of randomized trials shows that both impact and non-impact exercise have a positive effect at the lumbar spine in pre- and postmenopausal women. Impact exercise probably has a positive effect at the femoral neck. More studies are required to determine the optimal intensity and type of exercise.