This study was undertaken to follow up an incidence rate study of spinal injuries carried out from a cross-sectional random sample consisting of 2,342 workers who were compensated at least once in the year 1981 by the Quebec Worker's Compensation Board (QWCB) for an absence from work. These workers were followed for 3 years using the QWCB information system to record any recurrence of compensated absence from work. A total of 850 (36.3%) had at least one recurrence, and had longer episodes of absence than those without a recurrence (P less than 0.0001). Of these 850 workers, 824 (96.9%) had less than five recurrences and showed a systematic trend of gradual increasing duration of absence on each subsequent recurrence (P less than 0.05 in a repeated measures analysis of variance). A positive relationship was found between the duration of the initial episode of absence from work and the subsequent history of absence from work, both in terms of risk of recurrence (P less than 0.001) and of cumulated absence from work (P less than 0.0001), after controlling for age, sex, and site of symptoms. The computed instantaneous risk of entering a recurrence in a 31 year-old man experiencing lumbar symptoms, was 19.9% (95% confidence interval = 19.8-19.9) in the year following an initial episode of one day duration, and 26.7% (95% confidence interval = 24.3-29.3) after an initial episode of 6 months.