The support vector machine (SVM), as one of special regularization methods, has been used successfully in the field of pattern recognition. However, the traditional SVM, a supervised learning methed, gets the normal vector of the decision boundary mainly according to the largest interval principle but has not considered the underlying geometric structure and the discriminant information fully. Therefore, a global and local preserving based semi-supervised support vector machine: GLSSVM, is presented in this paper by introducing the basic theories of the locality preserving projections (LPP) and the within-class scatter of linear discriminant analysis (LDA) into the SVM. This method inherits the characteristics of the traditional SVM, fully considers the global and local geometric structure between samples, shows the global and local underlying discriminant information and meets the consistency assumption which the semi-supervised method must coincides with so that the shortcomings of the supervised methods can be overcome and the classification accuracy can be increased. The tests on the artificial and real datasets show the above mentioned advantages of the GLSSVM method.