The traditional linear discriminant analysis(LDA)algorithm does not consider the similarity between the samples embedded from the high-dimensional space to the low-dimensional space. Therefore, it often fails to achieve good results for non-Gaussian data. In this paper, a self-weight linear discriminant analysis algorithm is proposed. The new model assigns weights of the sample pairs by measuring the Euclidean distance between the sample pairs to differentiate the importance of each data point so that the underlying local manifold structure can be discovered, which can improve the ability of the model to process non- Gaussian data. Extensive experiments conducted on synthetic and real data demonstrate that the proposed algorithmhas improved the dimension reduction performance of traditional LDA in processing non-Gaussian data to some extent.