Occupational back injuries produced $27 billion in direct and indirect costs in 1988. Predictors of prolonged disability have generally been identified in selected clinical populations, but there have been few population-based studies using statewide registries from workers' compensation systems. This study uses a 1986 cohort of 8,628 Michigan workers with compensable back injuries followed to March 1, 1990. Cox proportional hazards analyses with nine categorical covariates identified factors predicting missed worktime for the first disability episode following the injury. The model distinguished factors affecting the acute (< or = 8 weeks) and chronic disability periods (> 8 weeks). The first disability episode following injury contains 69.6% of the missed worktime observed through follow-up. In the acute phase, which contributes 15.2% of first episode missed worktime, gender, age, number of dependents, industry (construction), occupation, and type of accident predict continued work disability. Marital status, weekly wage compensation rate, and establishment size do not. Beyond 8 weeks, age, establishment size and, to a lesser degree, wage compensation rate predict duration of work disability. Graphs show the predicted disability course for injured workers with specific covariate patterns. Future efforts to reduce missed worktime may require modifications in current clinical practice by patient age group and the development of new strategies to encourage small and medium-size employers to find ways to return their injured employees to work sooner. Recent federal statutes covering disabled workers will only partially correct the strong effect of employer establishment size.