Since falling is one of the main causes of morbidity and mortality among the elderly, attention should be paid to prevention. The purpose of the present study was to investigate the risk factors for falls among the elderly by a 5-year longitudinal observation. The subjects were the elderly aged 65 years and over living in a rural community in the eastern part of Honshu, Japan. Of the 852 eligible subjects, 685 persons (278 males and 407 females) received the multidimensional medical examination including physical performance as the baseline of TMIG-LISA (Tokyo Metropolitan Institute of Gerontology, Longitudinal and Interdisciplinary Study on Aging) in 1992. From baseline medical examinations, the following variables were adopted to investigate risk factors for falls (2 or more times during the 5-year follow-up): self-rated health, experience of falling, TMIG-index of competence, visual deficit, anthropometric measurements (height, weight, BML total skinfold thickness), lumbar bone mineral density, and physical performance tests (grip strength, finger-tapping rate, one leg standing time with eyes open or closed, preferred and maximum speed walking for 5 meters). A total of 108 subjects experienced falls two or more times during the 5-year follow-up period (43 men and 65 women). There were significant differences in grip strength, preferred and maximum walking speeds between those who fell and those who did not. Multiple logistic regression analysis controlled for age, sex and other variables adopted in this study revealed the following variables were significantly and independently related to falling: experience of falling within one year preceding the baseline survey (directly, p < 0.0001), preferred walking speed (inversely, p < 0.005) and total skinfold thickness (inversely, p < 0.01). These results suggest that the walking test as well as the experience of falling have discriminant and predictive validity for assessing the future occurrence of falls among the elderly.