Abstract Chernoff faces are a classical method for visualizing multidimensional data. This method represents multidimensional data in the shape of a human face. The motivation for using this approach is that humans are excellent at recognizing faces and noticing small changes in them. Thus, human visual judgment can be used to cluster the faces. However, different people may have different judgments of the same face, which will affect the analysis results. This paper presents a new approach to classify the faces. The main point of this method is to classify the Chernoff faces based on their quantified overall features. One Chernoff face is regarded as a geometric graphics group, and the distance between any two faces is calculated in the frequency domain via the V-system, which is a complete orthogonal function system on L 2[0,1]. The faces are classified by the resulting distance, so that we can obtain a uniform and reasonable evaluation result. Moreover, we can further evaluate the samples according to a given evaluation standard. This approach provides a new automated clustering method for Chernoff faces, which can avoid misjudgments due to human visual error. The experimental results indicate that the new method is simple, fast and effective. The classification result is the same as that obtained by SAS clustering in statistics.