Abstract: We present a method to perform Chinese dance motion retrieval based on pose space-time features according to the characteristics of Chinese dance movements.First,pose space-time features effectively reflect the locality,rhythm and subdivision of Chinese dance movements and provide a basis for performing Chinese dance motion retrieval.Second,an unsupervised feature selection algorithm for multivariate time series is introduced to remove redundant features and enhance the retrieval efficiency.Third,the relevance feedback based on feature weight adjustments is adopted to capture users’ intention and improve the retrieval performance.Experimental results show that this method has better retrieval performance and can perform Chinese dance motion retrieval effectively.