This paper presents a method for vision-based hand gesture recognition based on an approximate nearest neighbor search.The proposed method represents the shape of hands by the PCA-HOG descriptor,which is histogram of oriented gradients with principal component analysis based dimensionality reduction.This reduction can reduce the search time of finding an approximate nearest neighbor of a query image from a set of learned hand gestures.Experiments with sign-language images investigate the dimensionality reduction effect in terms of recognition rate.In addition,the experimental results show that the proposed method provides better performance than the SVM-based classifier.