OBJECTIVE: To investigate the predictive validity of fear avoidance beliefs as assessed by the Work Subscale (FABQ-W) of the Fear Avoidance Beliefs Questionnaire in a sample of 117 patients with a work-related musculoskeletal disorder, and identify two FABQ-W cut off points that identified participants as high or low risk of non return to work, following an interdisciplinary rehabilitation program. METHODS: A retrospective analysis of patient data collected in conjunction with the Victorian Workcover Authority "Sprains and Strains" program. Sequential logistic regression analysis was used to construct a model of prediction from the baseline variables of age, disability (using the Pain Disability Index), gender and FABQ-W scores. Receiver Operator Characteristic (ROC) curves were used to identify FABQ-W cut off points that best predicted the return to work outcome. RESULTS: Age and initial FABQ-W scores significantly improved the predictive capabilities of the model, but gender and disability did not. The model explained between 13.1% and 18.2% of the variability in the RTW outcome. ROC curves showed maximum sensitivity was 100% for a score of ≤ 27.5, with a score of > 39.5 identified as having optimum specificity (81.9%). CONCLUSION: Individuals with low FABQ-W scores are likely to return to work, however those with high scores will not necessarily have a poor outcome. This study supports the limited utility of the FABQ-W score for screening for risk of a poor return to work outcome in patients with a work related musculoskeletal disorder.