Purpose In the Netherlands, disability claimants are assessed after 2 years of sick leave, but their functioning may still improve. An accurate prognosis of functioning is difficult. Self predictions may be more accurate than those of professionals. The aim of this study, is to assess and compare the accuracy of predictions by disability claimants and insurance physicians (IPs) working at the Social Security Institute. It is further studied whether the accuracy differs between subgroups of claimants with mental or somatic health conditions. Methods We used data from the prospective cohort study cohort study. Following the assessment of the disability claim (n = 375) and after 1 year follow up (T1, n = 276) data on functioning were obtained from respondents by self-report questionnaire World Health Organization Disability Schedule 2.0. Both claimants and IPs were asked to predict improvement of functioning. Accuracy of their predictions were assessed by sensitivity, specificity, and area under the receiver operating curves (AUC). Mixed logistic regression was conducted to explore differences in accuracy between claimants with mental and somatic conditions. Results One-third (32 %) of disability claimants improved beyond the standard error of measurement. Disability claimants' and IPs were able to predict this improvement of functioning, but to a limited extent, with an AUC of 0.61 for IPs and 0.62 for disability claimants. We found no statistically significant differences in the accuracy of the predictions in claimants with mental or somatic health conditions. Conclusions Improvements of functioning were not uncommon. However, both IPs and disability claimants were unable to predict improvement with high levels of accuracy in both mental and somatic health conditions. © 2013 Springer Science+Business Media New York.