The least square support vector machine (LS-SVM) is used to predict the reliability of civil aircraft products based on small sample. The optimized input variable number of LS-SVM is determined through computing the saturated embedding dimension of reconstruct phase space. Then, a reliability prediction model is established by using LS-SVM and their parameters are also optimized by an automatic grid search method. The training and validation use reliability data from the hydraulic lock of a certain type aircraft. Finally, the one step and N-step prediction results of LS-SVM and radical basis function（RBF） neural network are compared, and show that the algorithm is feasible and valid for reliability prediction based on small sample.