Unsupervised feature selection algorithm based on manifold regularization self-representation MRSR directly constructed similarity matrix from the original sample space which might lead to inaccurate similarity description of samples in the reconstructed space. To solve this problem an unsupervised feature selection algorithm based on adaptive manifold regularization self-representation AMRSR was proposed. On the basis of MRSR algorithm unsupervised feature selection algorithm based on adaptive manifold regularization self-representation embedded the learning of similar matrix into the optimization process by imposing probabilistic nearest neighbor constraints on similar matrix and adaptively learned the similarity of samples in the reconstructed space so that more accurate local geometric manifold structure of samples could be obtained in each iteration and then representative features with local geometric manifold structure could be selected. Finally a large number of comparative experiments were carried out on four public datasets. By applying the feature selection results of the algorithms to K-means clustering two common clustering evaluation indexes were adopted clustering accuracy and normalized mutual information to evaluate the clustering effect. Experimental results show that the AMRSR algorithm has higher clustering accuracy and normalized mutual information than some existing algorithms which further indicates that the feature selection effect of this algorithm is better.