Public health decision making based on data sources that are characterized by a lack of independence and other complicating factors requires the development of innovative statistical techniques. Studies of injuries in occupational cohorts require methods to account for recurrent injuries to workers over time and the temporary removal of workers from the &risk set' while recuperating. In this study, the times until injury events are modelled in an occupational cohort of employees in a large power utility company where employees are susceptible to recurrent events. The injury history over a ten-year period is used to compare the hazards of speci"c jobs, adjusted for age when "rst hired, and race/ethnicity di!erences. Subject-speci"c random e!ects and multiple event-times are accommodated through the application of frailty models which characterize the dependence of recurrent events over time. The counting process formulation of the proportional hazards regression model is used to estimate the e!ects of covariates for subjects with discontinuous intervals of risk. In this application, subjects are not at risk of injury during recovery periods or other illness, changes in jobs, or other reasons. Previous applications of proportional hazards regression in frailty models have not needed to account for the changing composition of the risk set which is required to adequately model occupational injury data.