Data concerning all accidents with absence from work exceeding three days were analysed in a cohort of 5000 workers (all employees of a Local Health Unit). A total of 1062 accidents were recorded in the file from 1987 to 1989, 118 coded as "low-back injury". A case-control study design was used to detect the main causes of these accidents, using accidents coded as other than low-back injury as controls. The odds ratios and related confidence limits were calculated in order to assess the degree of association, taking account of potentially confounding factors. A significant relationship was observed between hospital wards and other health service departments rated as "high exposure" and low-back injury (OR high exposure versus low exposure = 2.13; 95% C.L. 1.06-4.29). Selection bias and its influence on the results is also discussed.