This study assesses risk factors for sickness absence in the British civil service. As one component of a large study of psychosocial factors and health, 10,314 civil servants between the age of 35 and 55 completed questionnaires about their work environment, social circumstances outside work, health and health-related behaviours. To obtain a more objective measure of the work environment, personnel managers provided external assessments of participants' jobs. The baseline variables were related to rates of short spells (7 days or less) and long spells (more than 7 days) of sickness absence for 85% of participants, over a mean period of 20 months (6-26 months). There were striking grade differences in sickness absence, with a stepwise increase in rates of both short and long spells from top administrators to clerical and office support staff. Other identified risk factors explained only a third of these grade differences in sickness absence. Further analyses were adjusted for age and grade. Self-reported health was strongly related to rates of long spells and, to a lesser extent, short spells. Adequacy of support and difficulty paying bills were the two factors outside work which related to rates of both short and long spells. Job satisfaction was the only measure of the work environment which related to rates of both short and long spells. Other aspects of the work environment which were associated with increased rates of short spells were low variety and skill use and low support at work for both sexes, and low control, slow work pace and few conflicting demands for men. Self-reports and external assessments of the work environment related to sickness absence in a similar way, suggesting that the work environment itself was important. Factors which did not relate to either short or long spells of sickness absence were marital status, dependent children, the frequency of social contacts and physical activity. Women had higher rates of sickness absence than men and Asians had higher rates than Caucasians. This study identified a number of risk factors for sickness absence and differences in these risk factors for short and long spells of sickness absence. The grade, sex and ethnic differences in sickness absence remained largely unexplained. Group attitudes towards sickness absence may be important. Methodological issues related to the assessment of psychosocial factors are discussed.