Background: Knowledge on predictors of disability pension is very limited. The aim was to assess the importance of sick-leave diagnosis and socio-demographic variables as risk factors for disability pension among individuals on long-term sickness absence and to compare these factors by gender and over time. Methods: A prospective population-based cohort study in Östergötland County, Sweden, included 19 379 individuals who, in 1985-87, were aged 16-60 years and had a new spell of long-term sickness absence lasting <56 days. Follow-up was done in two time frames: 0-5 and 6-10 years after inclusion. The risk of disability pension in relation to sick-leave diagnosis and socio-demographic factors was assessed by Cox proportional hazard regression analysis. Results: In 5 years, after inclusion, 28% of the cohort had been granted disability pension. Those with higher age, low income, previous sick leave, no employment and non-Swedish origin had higher risk of disability pension, while those with young children had lower risk. Considering the inclusion diagnosis, the pattern differed between men and women (P < 0.001). Among men, those with mental disorders had the highest risk and among women those with musculoskeletal disorders. Except for income, the effect of which was reversed over time, the overall pattern of disability pension predictors remained 6-10 years after inclusion but was attenuated. Conclusion: Besides socio-demographic risk factors, the sick-leave diagnoses constitute an important both medium and long-term predictor of disability pension among both men and women on long-term sickness absence. © 2008. The Author(s).