The most of unsupervised manifold learning algorithms provide the embedding only for the given training points, with no straightforward extension for incremental points. This paper provides an incremental local tangent space alignment(LTSA) algorithm based on selecting landmark points for the clustering of incremental points. Firstly, the LTSA algorithm is used to construct the global coordinate matrix for the given training points. The landmark points of the given training points are select by the least angle regression(LARS) algorithm. Using the landmark points and given training points construct the global coordinate matrix for all points, and the lower-dimensional embedding coordinate of the incremental points are estimated by the global coordinate matrix and lower-dimensional embedding coordinate of the given training points. Then the lower-dimensional embedding coordinate of all points are updated by the eigenvalue iteration algorithm. Experiments with the proposed method are carried out to identify the four different mode of the vibration signal of rolling bearing. The experiments result demonstrates the incremental LTSA algorithm based on selecting landmark points can preserve the class separability measurement of different sorts on the basis of straightforward incremental learning for out-of-sample.