The increase in care-receiver/caregiver ratio is becoming a social problem in Japan, one factor in which is the increase in the number of falls in the elderly. The purpose of this study is to develop a balance training ability system for fall prevention. To effectively improve balance ability while walking, dynamic training specialized for somatic sensation is useful. We developed a system that greatly improves balance ability by removing visual information of the outside world during gait training and by producing visual feedback of somatic sensation information, consisting of joint angle information about the whole body and foot pressure distribution information, that is, somatic sensation information used for gait. One technical problem with this system is the selection of display method at the time of visual feedback. Therefore, in this paper, we hypothesize that the training effect improves when a combination of foot pressure information and joint information is used. As specific presentation information, center of pressure (hereinafter called CoP) and ideal CoP traces derived from each joint angle were used. The ideal CoP trace is shown with a broken line connecting the heel, metatarsal and toe-tip part of the foot, and is updated every 10 walking cycles by changing the transverse coordinates of the metatarsal and toe-tip. To verify the abovementioned hypothesis, we created a machine equipped with the feedback system, and verified and tested it. To create the feedback system, the relationship between each joint angle in the body and ideal CoP traces were derived by a multiple regression analysis of the gait data (foot pressure distribution, joint angle) of young, physically unimpaired volunteers. In the verification experiment, we compared the training using the feedback system to that without in a visual sense cutoff state. The standard deviation a of right-and-left directions of the CoP traces was used as a performance index of balance ability. Increased a indicates reduced dynamic balance ability. We found that a increased by 0.61 times by feedback training, which was a significant decrease, and by 0.84 times after the visual sense interception training. A significant difference (p < 0.05) was seen between the effects of these two training types, suggesting that dynamic balance ability is improved by our feedback system. In future, we will verify the long-term training effects over several months or years, as well as the improvement in motivation over training.