BACKGROUND: The interpretation of the change scores of the Barthel Index (BI) in follow-up or outcome studies has been hampered by the fact that its minimal clinically important difference (MCID) has not been determined. OBJECTIVE: This article was written to establish the MCID of the BI in stroke patients. METHODS: Both anchor-based and distribution-based methods were used to establish the MCID. In the anchor-based method, 43 stroke inpatients participated in a follow-up study designed to determine the MCID of the BI using patients’ global ratings of the activities of daily living function on a 15-point Likert-type scale. The mean change scores on the 20-point scale of the BI of the MCID group, based on the patients’ ratings on the Likert-type scale, served as the first estimate of the MCID. In the distribution-based method, 56 chronic stroke patients participated in the test-retest reliability study to determine the MCID of the BI. One standard error of measurement (SEM) served as the second estimate for the MCID. The larger MCID value of the 2 estimates was chosen as the MCID of the BI. RESULTS: In the anchor-based study, there were 20 patients in the MCID group, with a mean change score of 1.85 points (ie, the first MCID estimate). In the distribution-based study, the SEM based on test-retest agreement was 1.45 points (ie, the second MCID estimate). The MCID of the BI in stroke patients was estimated to be 1.85 points. CONCLUSION: The authors’ results, within the limitations of their design, suggest that if the mean BI change score within a stroke group has reached 1.85 points in a study, the change score on the BI can be perceived by patients as important and beyond measurement error (ie, such a change score is clinically important).