Abstract: It is difficult for methods based on word spaces to handle with the high dimensionality characteristic and complex correlation of the texts vectors.To solve this problem, a algorithm based on kernel principal component analysis and RBF neural network is proposed.First, this new algorithm employs KPCA with a appropriate kernel function to find the principal components of the input vectors in the high dimensional feature space, which effectively reduces the dimensionality of input vectors and gets the semantic feature space.Then, we train a RBF neural network in the semantic feature space.The experiment results show that the new method can effectively reduce the dimensionality of the data sets and notably enhance the classification precision while reduces the training time of the RBF networks.