The Physical Activity Questionnaire for Elderly Japanese (PAQ-EJ) is a self-administered physical activity questionnaire for elderly Japanese; the authors report here on its repeatability and direct and indirect validity. Reliability was assessed by repeat administration after 1 month. Direct validation was based on accelerometer data collected every 4 s for 1 month in 147 individuals age 65-85 years. Indirect validation against a 10-item Barthel index (activities of daily living [ADL]) was completed in 3,084 individuals age 65-99 years. The test-retest coefficient was high (r = .64-.71). Total and subtotal scores for lower (transportation, housework, and labor) and higher intensity activities (exercise/sports) were significantly correlated with step counts and durations of physical activity <3 and >or=3 METs (r = .41, .28, .53), respectively. Controlling for age and ADL, scores for transportation, exercise/sports, and labor were greater in men, but women performed more housework. Sex- and ADL- or age-adjusted PAQ-EJ scores were significantly lower in older and dependent people. PAQ-EJ repeatability and validity seem comparable to those of instruments used in Western epidemiological studies.