Questions: Can clinical prediction rules be derived for injured Australian workers with persistent musculoskeletal pain? Are they valid? Design: Longitudinal observational study. Participants: 847 injured workers with persistent musculoskeletal pain undergoing rehabilitation. Outcome measures: At baseline, 12 putative predictors were measured. At 9 weeks, short-term outcomes such as pain (visual analogue scale), activity limitation (Functional Rating Index) and work upgrade (increase in work hours or duties) were measured. At 6 months, long-term work status (working or not working) was measured. Results: Data were obtained from 85% of the participants who were followed up at both 9 weeks (720 of 847) and 6 months (247 of 290). Predictors of outcome included high baseline pain and activity limitation, long duration of previous intervention, not working, non-English speaking background, and the area of pain. Accuracy was highest for clinical prediction rules predicting pain and level of activity limitation at 9 weeks (R 2 = 0.67 and 0.69 respectively) and work status at 6 months (LR- = 0.24). Conclusion: Accurate clinical prediction rules have been derived and validated for injured workers with persistent musculoskeletal pain, predicting activity limitation, pain, and work outcomes following exercise-based rehabilitation. Further research to validate these prediction rules in other populations and to assess the effectiveness of tailoring intervention based on the estimated prognosis would be valuable.