OBJECTIVE: To investigate the validity of item score summation for the original and modified versions of the Barthel Index. DESIGN: Rasch analysis of Barthel Index data. SETTING: General medical wards at 2 acute care hospitals in Australia. PARTICIPANTS: Consecutive older medical patients (N=396). INTERVENTIONS: Not applicable. MAIN OUTCOME MEASURES: Activity limitation was assessed by using the Barthel Index at hospital admission and discharge. At 1 hospital site, the original Barthel Index was used, and at the other hospital site the Modified Barthel Index (MBI) was used. RESULTS: More than half of the items showed misfit to the Rasch model for both versions of the Barthel Index. The continence items appear to measure a different construct to the other items. After the removal of the continence items, data for the remaining items still did not fit the Rasch model. Neither the original nor the MBI are unidimensional scales. An exception to this occurred when the original Barthel Index was rescored and only then for discharge and not for admission Barthel Index data. CONCLUSIONS: Because clinicians do not typically rescore outcomes obtained by using the Barthel Index, these findings, combined with unacceptable ceiling effects, render the Barthel Index an assessment tool with limited validity for measuring and monitoring the health of older medical patients.