The research of sleep staging is an important basis of evaluating sleep quality and diagnosing diseases. In order to achieve automatic sleep staging, we proposed a new method which combines with principal component analysis (PCA) and support vector machine (SVM) for automatic sleep staging. Firstly, we used PCA to reduce dimension of time-frequency-space domains and nonlinear dynamical characteristics of sleep EEG from 5 subjects to reduce data redundancy. Secondly, we used 1-a-1 SVM to classify sleep stages. The results showed that the correct rate can reach 89.9%, which was better than those of many other similar studies.