OBJECTIVES: Condition-specific preference indices (COPI's) may provide a sensitive measure of treatment effect and economic impact in certain disorders. Our goal was to adapt the Barthel Index, a validated 10-item measure of physical disability, as a COPI for stroke clinical trials and practice settings. The resulting instrument is known as the Barthel Preference Index (BPI). METHODS: Preferences were elicited from 152 community members (≥45 years) using visual analog scaling (VAS) and discrete choice experiment (DCE). Ordinary least squares and probit regression models were developed from VAS and DCE data, respectively, and indexed from zero to one. Data from stroke inpatients were used to calculate preferences for commonly occurring stroke disability states. Tests of internal consistency, and construct, convergent and discriminant validity were performed. RESULTS: Thirty-one percent of the subjects were male, mean age was 61 years, 88% were Caucasian, 60% had more than a high school education, and 63% were married. VAS preferences ranged from 0.09 to 0.40 for severe and very severe stroke, 0.40 and 0.74 for moderate stroke, and 0.72 to 0.97 for mild stroke. DCE preferences ranged from 0.04 to 0.17, 0.17 to 0.76, and 0.47 to 0.88, respectively. Multiple preferences were observed within individual BI scores. Both models demonstrated internal consistency, and construct and convergent validity. Spearman correlations between VAS and DCE utilities, and BI scores were 96% and 91%, respectively. The correlation between VAS and DCE utilities was 94%. Overall, the VAS model was stronger and demonstrated superior discriminant validity. CONCLUSIONS: The Barthel Preference Index, a COPI intended for stroke clinical trials and practice, was found to be valid and reliable in this population. Multiple preferences within scores suggest that the BPI may be more sensitive to treatment effect than BI scores. Further validation and tests of responsiveness in clinical trials are required.