OBJECTIVES: To assess the performance of the EQ-5D-5L Japanese version compared with the EQ-5D-3L in clinical setting targeted at patients with stroke in Japan. METHODS: The stroke patients were recruited from six rehabilitation hospitals in Japan. The proxy of the patients completed a questionnaire including the two versions of the EQ-5D. We tested the inconsistency, the redistribution properties, and the ceiling effect. RESULTS: A total 533 patients were recruited: 61% male, 15-99 (mean=67) years old. Diagnoses of patients were infarction (54.6%), hemorrhage (35.5%), and other (9.9%). The proportion of inconsistent responses (i.e., 3L responses that were at least two levels away from the 5L responses) was 3.6%. In particular the proportion in MOBILTY was 14.6%. Regarding redistribution, 52-61% of the patients answering level 2 with the 3L version redistributed their responses to levels 2 or 4 with the 5L version. A relative 3% reduction of the ceiling effect was found. CONCLUSIONS: Our findings suggest that the EQ-5D-5L Japanese version performs better in at least some properties analyzed. Further study is necessary to clarify other psychometric properties.