Objective. Examine the effects of exercise on femoral neck (FN) and lumbar spine (LS) bone mineral density (BMD) in premenopausal women. Methods. Meta-analysis of randomized controlled exercise trials ≥24 weeks in premenopausal women. Standardized effect sizes (g) were calculated for each result and pooled using random-effects models, Z score alpha values, 95% confidence intervals (CIs), and number needed to treat (NNT). Heterogeneity was examined using Q and I(2). Moderator and predictor analyses using mixed-effects ANOVA and simple metaregression were conducted. Statistical significance was set at P ≤ 0.05. Results. Statistically significant improvements were found for both FN (7g's, 466 participants, g = 0.342, 95% CI = 0.132, 0.553, P = 0.001, Q = 10.8, P = 0.22, I(2) = 25.7%, NNT = 5) and LS (6g's, 402 participants, g = 0.201, 95% CI = 0.009, 0.394, P = 0.04, Q = 3.3, P = 0.65, I(2) = 0%, NNT = 9) BMD. A trend for greater benefits in FN BMD was observed for studies published in countries other than the United States and for those who participated in home versus facility-based exercise. Statistically significant, or a trend for statistically significant, associations were observed for 7 different moderators and predictors, 6 for FN BMD and 1 for LS BMD. Conclusions. Exercise benefits FN and LS BMD in premenopausal women. The observed moderators and predictors deserve further investigation in well-designed randomized controlled trials.