OBJECTIVES: This study aimed to explore and compare the ability of five instruments for self-rating to predict future sick leave rates. METHODS: In three Swedish municipalities 2,252 employees completed a baseline questionnaire and were followed up for 4 years. Five health-oriented instruments for self-rating were used as potential predictors of the two outcome measures no sick leave at all, and one or more spells of long-term sick leave >/=28 days. Positive and negative predictive values as well as Cox proportional hazard ratios (denoted as RRs) adjusted for age and work type were calculated. RESULTS: The instruments showed no statistical difference in predicting future sick leave for either of the sexes. For no sick leave RRs ranged between 1.27 and 1.52 (women), 1.35 and 1.61 (men); for long-term sick leave RRs ranged between 1.78 and 2.39 (women), 2.87 and 5.53 (men). However, the best prediction of long-term sick leave for men, RR 5.53, 95% confidence interval (CI) 3.37-9.08, was significantly higher than the best prediction for women, RR 2.39, 95% CI 1.97-2.90. CONCLUSION: Prediction of long-term sick leave was better than that of no sick leave, and better among men than among women. There was a tendency for somewhat better prediction of future sick leave by multiple-question instruments, but single-question instruments can very well be used in predicting future sick leaves, and crude analyses stratified by sex can be used for screening purposes.