BackgroundLow back pain is a major health concern in China because of its high prevalence among the general population. Exercise is universally recommended as a core treatment to reduce the symptoms, but the relative efficacy of different types of exercise have not been evaluated, and the most appropriate exercise intervention is still unknown. We therefore aimed to compare and rank the effectiveness of different exercise interventions in relieving pain and improving function in patients with low back pain using a network meta-analysis. MethodsWe carried out a systematic review and network meta-analysis. We searched PubMed, the Cochrane Central Register of Controlled Trials, Embase, and the reference lists of relevant articles for randomised controlled trials (RCT) published up to May 15, 2018. We included RCT studies comparing exercise interventions with each other or with controls (usual care or no exercise), for adults with low back pain. We excluded non-RCT studies and studies that did not investigate exercise interventions. The primary outcomes were improvement in pain intensity (as measured by the numerical rating scale, the visual analogue scale, or others) and disability (as measured by the Morris disability index, the Oswestry disability index, or others). We used the standardised mean difference (SMD) and 95% CI to measure effect sizes, and did pairwise meta-analysis and network meta-analysis to combine both direct (within trial) and indirect (between trial) evidence of exercise effectiveness. The quality of evidence was assessed using the GRADE framework. FindingsWe identified 5167 relevant studies, and of these we included 41 trials involving 3050 participants. 22 types of exercise interventions for low back pain published until May 15, 2018 were included. In terms of pain relief, aerobic, extension, stabilisation, strengthening, and stretching exercise (alone, not in combination) were significantly more effective than usual exercise, with SMD ranging between −2.19 (95% CI −4·25 to −0·13, extension exercise) and −0·71 (–1·37 to −0·05, stretching exercise). Aerobic strengthening, aerobic strengthening plus stretching, aquatic aerobics and any combination of stabilisation, strengthening, and stretching exercise were also more efficacious compared with usual exercise. When all interventions were considered, stabilisation plus strengthening exercise (SMD −4·27, 95% CI −6·03 to −2·51; P-score, 96% likelihood) was ranked the most effective for pain relief, followed by flexion (–2·38, −4·81 to 0·05; 78%), extension (–2·19, −4·25 to −0·13; 76%), and aquatic aerobic exercise (–1·76, −2·88 to −0·63; 72%). Regarding disability, only aerobic (–0·68, −1·22 to −0·14), stabilisation (–1·15, −2·17 to −0·12), aerobic strengthening (–1·06, −2·09 to −0·03), aquatic aerobic (–1·21, −2·32 to −0·09), stabilisation plus strengthening (–2·88, −4·61 to −1·14), strengthening plus stretching (–1·56, −2·88 to −0·24), and strengthening plus stretching plus stabilisation exercises (–1·66, −2·45 to −0·88) were more efficacious than usual exercise. Of all of the studied interventions, a combined intervention of stabilisation and strengthening exercise showed the greatest effect on disability, with a P-score of 92%. The strength of evidence according to the GRADE assessment was low to moderate. InterpretationAn approach combining stabilisation and strengthening exercises is likely to be most effective in the management of low back pain, but most individual exercise interventions alone and various combinations of interventions also had significant beneficial effects compared with usual exercise. These results should serve evidence-based practice and guide patients and caregivers on the choice of intervention. FundingNational Natural Science Foundation of China (81371942), and Doctoral Scientific Research Foundation of Liaoning Province, China (201601304).