Background Fear of pain and movement is an important factor in the development of hypervigilance and avoidance behaviours. Objective We examined the efectiveness of exercise training on improving fear-avoidance beliefs. Methods A systematic review (data sources: MEDLINE, CINAHL, SPORTDiscus, EMBASE, CENTRAL) and metaanalysis of randomised controlled/clinical trials of exercise training in adults versus relevant nonexercise comparators that quantifed fear-avoidance was conducted. Results After screening 4603 identifed records, 17 (2014 participants) and 13 (1152 participants) studies were eligible for qualitative and quantitative synthesis, respectively. Pairwise meta-analysis showed exercise training was more efective than all non-exercise comparators (standardised mean diference (SMD) [95% CI] − 0.378 [− 0.623, − 0.133], P=0.002, Grading of Recommendations Assessment, Development and Evaluation [GRADE]: very low) for reducing fear-avoidance. Exercise training was more efective than true control for reducing fear avoidance (− 0.407 [− 0.750, − 0.065], P=0.020, GRADE: very low), however it was not more efective than other interventions (− 0.243 [− 0.614, 0.128], P=0.199, GRADE: very low). In people with low back pain, exercise training was more efective than non-exercise comparator groups for reducing fear-avoidance (− 0.530 [− 0.755, − 0.304], P<0.001, GRADE: very low). For individuals with neck pain, exercise training was not more efective than non-exercise comparator groups for reducing fear-avoidance (0.061 [− 0.360, 0.482], P=0.777, GRADE: very low). Conclusion There is very low to low-quality evidence that exercise training is efective for reducing fear-avoidance, including in people with low back pain. Exercise training may be more efective than no intervention for reducing fear avoidance, but there is very low-quality evidence that non-exercise interventions are as efective as exercise for fear avoidance. Few studies with low risk of bias is a limitation.