Abstract Objective Effective use of rehabilitation facilities for older people requires that those selected to undergo rehabilitation are best placed to benefit. We tested whether routinely collected clinical factors could predict deterioration or failure to improve during inpatient rehabilitation. Methods Analysis of prospectively collected routine clinical data from adults aged 65 and over, admitted to an inpatient rehabilitation between 1st January 1999 and 31st December 2008. Measures analysed were changed in 20-point Barthel score and indices of function including nutrition, swallow, communication and mental health between admission and discharge. Cut-off values for admission Barthel score were used to test which groups of patients would fail to improve further; admission indices of function, comorbidity and demographic information were combined in multivariable analyses to test which factors independently predicted failure to improve, deterioration or death during inpatient rehabilitation. Results Three thousand five hundred and seventy-two patients were included in the analyses, mean age 81.6 (SD 7.6) years. The mean admission Barthel score was 10.2 (SD 3.8). There was no admission Barthel score above which patients failed to improve either their Barthel score or other indices of function. In multivariate analyses, combinations of age, sex, admission Barthel score and other admission indices of function independently predicted inpatient death, death or deterioration, and death, deterioration or failure to improve on admission Barthel. The classification accuracy for all models was low at 70% or less. Conclusions Barthel scores and indices of function do not accurately predict which older patients will fail to benefit from inpatient rehabilitation.