Background Low back pain (LBP) is responsible for considerable personal suffering worldwide. Those with persistent disabling symptoms also contribute to substantial costs to society via healthcare expenditure and reduced work productivity. While there are many treatment options, none are universally endorsed. The idea that chronic LBP is a condition best understood with reference to an interaction of physical, psychological and social influences, the 'biopsychosocial model', has received increasing acceptance. This has led to the development of multidisciplinary biopsychosocial rehabilitation (MBR) programs that target factors from the different domains, administered by healthcare professionals from different backgrounds. Objectives To review the evidence on the effectiveness of MBR for patients with chronic LBP. The focus was on comparisons with usual care and with physical treatments measuring outcomes of pain, disability and work status, particularly in the long term. Search methods We searched the CENTRAL, MEDLINE, EMBASE, PsycINFO and CINAHL databases in January and March 2014 together with carrying out handsearches of the reference lists of included and related studies, forward citation tracking of included studies and screening of studies excluded in the previous version of this review. Selection criteria All studies identified in the searches were screened independently by two review authors; disagreements regarding inclusion were resolved by consensus. The inclusion criteria were published randomised controlled trials (RCTs) that included adults with non-specific LBP of longer than 12 weeks duration; the index intervention targeted at least two of physical, psychological and social or work-related factors; and the index intervention was delivered by clinicians from at least two different professional backgrounds. Data collection and analysis Two review authors extracted and checked information to describe the included studies, assessed risk of bias and performed the analyses. We used the Cochrane risk of bias tool to describe the methodological quality. The primary outcomes were pain, disability and work status, divided into the short, medium and long term. Secondary outcomes were psychological functioning (for example depression, anxiety, catastrophising), healthcare service utilisation, quality of life and adverse events. We categorised the control interventions as usual care, physical treatment, surgery, or wait list for surgery in separate meta-analyses. The first two comparisons formed our primary focus. We performed meta-analyses using random-effects models and assessed the quality of evidence using the GRADE method. We performed sensitivity analyses to assess the influence of the methodological quality, and subgroup analyses to investigate the influence of baseline symptom severity and intervention intensity. Main results From 6168 studies identified in the searches, 41 RCTs with a total of 6858 participants were included. Methodological quality ratings ranged from 1 to 9 out 12, and 13 of the 41 included studies were assessed as low risk of bias. Pooled estimates from 16 RCTs provided moderate to low quality evidence that MBR is more effective than usual care in reducing pain and disability, with standardised mean differences (SMDs) in the long term of 0.21 (95% CI 0.04 to 0.37) and 0.23 (95% CI 0.06 to 0.4) respectively. The range across all time points equated to approximately 0.5 to 1.4 units on a 0 to 10 numerical rating scale for pain and 1.4 to 2.5 points on the Roland Morris disability scale (0 to 24). There was moderate to low quality evidence of no difference on work outcomes (odds ratio (OR) at long term 1.04, 95% CI 0.73 to 1.47). Pooled estimates from 19 RCTs provided moderate to low quality evidence that MBR was more effective than physical treatment for pain and disability with SMDs in the long term of 0.51 (95% CI -0.01 to 1.04) and 0.68 (95% CI 0.16 to 1.19) respectively. Across all time points this translated to approximately 0.6 to 1.2 units on the pain scale and 1.2 to 4.0 points on the Roland Morris scale. There was moderate to low quality evidence of an effect on work outcomes (OR at long term 1.87, 95% CI 1.39 to 2.53). There was insufficient evidence to assess whether MBR interventions were associated with more adverse events than usual care or physical interventions. Sensitivity analyses did not suggest that the pooled estimates were unduly influenced by the results from low quality studies. Subgroup analyses were inconclusive regarding the influence of baseline symptom severity and intervention intensity. Authors' conclusions Patients with chronic LBP receiving MBR are likely to experience less pain and disability than those receiving usual care or a physical treatment. MBR also has a positive influence on work status compared to physical treatment. Effects are of a modest magnitude and should be balanced against the time and resource requirements of MBR programs. More intensive interventions were not responsible for effects that were substantially different to those of less intensive interventions. While we were not able to determine if symptom intensity at presentation influenced the likelihood of success, it seems appropriate that only those people with indicators of significant psychosocial impact are referred to MBR.