The dimensionality reduction has always been a long lasting thorny problem on the study of facial expression recognition (FER). In this paper, we propose a novel method of facial expression recognition, which using Gaussian process latent variable models (GPLVM) for reducing the high dimensional data of facial expression images into a relatively low dimension data and using support vector machine (SVM) classifier for the expression classification lately. By applying this algorithm to Japanese Female Facial Expression (JAFFE) database for facial expression recognition, we find that the proposed new algorithm has a better performance than the traditional algorithms, such as PCA and LDA etc. This have further proved the effectiveness of our proposed algorithm.