In this paper, we focus on the performance of Harris feature point extraction method (Harris Method) and evaluate its characteristic in some aspects. We propose a set of criteria for the evaluation: Detection, Accuracy, Consistency, and Efficiency. All of these criteria are defined theoretically and measured with different kinds of images. Especially, to verify the localization accuracy of Harris Method, we synthesize a series of chessboard images in which the real coordinates of feature points could be computed. Therefore, the criteria proposed here are objective and believable.