Recently, some medical institutions introduce a walking assist robot to improve an efficiency of rehabilitation for patients with hemiplegia. However, the new rehabilitation program with walking assist robot for the patients is not established. Therefore, we define new rehabilitation skills as the rehabilitation methodology to apply the walking assist robot to the patients. By interviewing the expert, we found that the expert classifies the types of the patients by their gait patterns in order to use the walking assist robot correctly. However, it is difficult for non- expert to classify these patterns because this classification skill is still a tacit knowledge. Therefore, our objective is to develop an auxiliary system to classify the types of patients with hemiplegia for transferring skill of rehabilitation for using the walking assist robot properly by translating the explicit knowledge from the tacit knowledge. First, we extracted the skill of the expert by interviewing. Second, we extracted the gait features by observing the movies of the patients’ gait. The movies of the patients’ gait which are labelled in advance by the expert were used. Then, we arranged the gait features of the difference points due to the type of the patients. In this study, sixteen movies, including four types of the patients, were used. As a result of the interview, the patterns are composed of four types which the expert are defined by the timing of gait event. As a result of the observation of the movies, gait stride, duration of the stance phase, and walking rate were suitable for the criteria of the classification. This rule is correspond with the expert’s policy.