OBJECTIVES--To determine which factors measured at baseline and during the course of treatment influence time to return to work after a first compensated episode of back pain. METHODS--The design is a treatment inception cohort including 305 compensated workers out of 402 eligible ones presenting at two rehabilitation centres for conventional treatment. Crude and adjusted rate ratios (RRs) along with 95% confidence intervals (95% CIs) were estimated with the Cox's proportional hazards regression. RESULTS--50% of workers had not returned to work after 112 days of follow up, and 11.3% still had not after 270 days. At the end of the study period (maximum follow up time was 1228 days), 230 workers (75.4%) had returned to work, 6.5% had not, and a similar percentage had retired, gone into vocational training, or returned to school. In the final model stratified for radiating pain during treatment, which was an important prognostic variable, workers between 21 to 30 years of age had a greater chance of returning to work (RR (95% CI) 1.43 (1.04 to 1.98) than those > or = 30. The other factors associated with a greater chance of returning to work were: a diagnosis of sprain or pain upsilon a diagnosis of intervertebral disc disorder (2.20 (1.23-3.91)), < 30 days of waiting between the accident and the beginning of treatment (1.30 (0.96 to 1.77)), a good flexion at baseline (1.52 (1.04 to 2.23)), absence of neurological symptoms during treatment (1.40 (0.98-2.00)), > 24 months of employment in the industry (1.49 (1.10 to 2.03)), working for a public industry upsilon a private one (1.63 (1.21 to 2.19)), and the ability to take unscheduled breaks (1.45 (1.06 to 1.97)). CONCLUSIONS--Even with a first episode of back pain, time to return to work is long and the proportion not returning is high. Return to work as expected is influenced by disease and host characteristics but also by social and work factors. Reinstatement programmes should account for all these factors.