OBJECTIVE: To compare the test-retest reliability and random measurement errors of the Barthel Index (BI) and modified Barthel Index (MBI) in patients with chronic stroke. METHOD: The intraclass correlation coefficient (ICC) and the minimal detectable change (MDC) were applied respectively to examine the test-retest reliability (about 2 weeks apart) and the random measurement errors. The MDC% was used to adjust the cut-off score for determining whether a real change had been achieved, if heteroscedasticity existed. RESULTS: A total of 60 patients participated. The BI and MBI both had high ICCs (0.94 and 0.94, respectively) with small MDCs (16.2 and 15.4, respectively) and MDC%s (21.2% and 19.0%, respectively), indicating that both measures have comparable reliability in repeated assessments. However, moderate associations ( = -0.47 for the BI and -0.59 for the MBI) were found between the means of tests and retests and the absolute values of change scores, indicating heteroscedasticity. These findings suggest that a fixed MDC value is not appropriate for determining the real change in both measures because the amount of random measurement error varies with the patients' ADL function. CONCLUSION: The MBI, which showed excellent test-retest reliability and relatively lower random measurement error than the BI, appears to be a better ADL measure. The MDC% adjusted value is recommended to determine whether the change scores are beyond random measurement error.IMPLICATIONS FOR REHABILITATIONThe MBI is recommended for clinical and research applications because it has better test-retest reliability and relatively lower random measurement error than those of the original BI.The MDC% adjusted value is recommended to determine whether the change scores are beyond random measurement error when the MBI or the BI is used.