Pain is the most disabling characteristic of musculoskeletal disorders, and while exercise is promoted as an important treatment modality for chronic musculoskeletal conditions, the relative contribution of the specific effects of exercise training, placebo effects and non-specific effects such as natural history are not clear. The aim of this systematic review and meta-analysis was to determine the relative contribution of these factors to better understand the true effect of exercise training for reducing pain in chronic primary musculoskeletal pain conditions. Systematic review with meta-analysis DATA SOURCES: MEDLINE, CINAHL, SPORTDiscus, EMBASE and CENTRAL from inception to February 2021. Reference lists of prior systematic reviews. Randomised controlled trials of interventions that used exercise training compared to placebo, true control or usual care in adults with chronic primary musculoskeletal pain. The review was registered prospectively with PROSPERO (CRD42019141096). We identified 79 eligible trials for quantitative analysis. Pairwise meta-analysis showed very low-quality evidence (GRADE criteria) that exercise training was not more effective than placebo (g [95% CI]: 0.94 [-0.17, 2.06], P = 0.098, I2 = 92.46%, studies: n = 4). Exercise training was more effective than true, no intervention controls (g [95% CI]: 0.99 [0.66, 1.32], P<0.001, I2=92.43%, studies: n=42), usual care controls (g [95% CI]: 0.64 [0.44, 0.83], P<0.001, I2=76.52%, studies: n=33), and when all controls combined (g [95% CI]: 0.84 [0.64, 1.04], P < 0.001, I2 = 90.02%, studies: n = 79). There is very low-quality evidence that exercise training is not more effective than non-exercise placebo treatments in chronic pain. Exercise training and the associated clinical encounter are more effective than true control or standard medical care for reductions in pain for adults with chronic musculoskeletal pain, with very low quality of evidence based on GRADE criteria.