In this paper, followed the assumption that the gene expression data of tumor may be sampled from the data with a probability distribution on a sub-manifold of ambient space, a supervised version of locally linear embedding (LLE), named locally linear discriminant embedding (LLDE), is proposed for tumor classification. In the proposed algorithm, we construct a vector translation and distance rescaling model to enhance the recognition ability of the original LLE from two aspects. To validate the efficiency, the proposed method is applied to classify two different DNA microarray datasets. The prediction results show that our method is efficient and feasible.