Existing ensemble clustering approaches usually construct a co-association matrix by combining clustering members and then use it to achieve the clustering result. However, the co-association matrix is easily influenced by low-quality clustering members, leading to poor clustering results. To address the problem, we propose a three-layer weighted ensemble clustering approach based on tensor decomposition. Our approach firstly computes weights for the partition, cluster, and point layers using the normalized mutual information, the information entropy, and the quadratic function, respectively, to reduce the influence of low-quality clustering members. Next, we construct a coherent-link matrix and a three-layer weighted co-association matrix and stack them into a 3-dimensional tensor. Finally, we leverage the tensor’s low-rank structure to propagate reliable information from the coherent-link matrix, refining the three-layer weighted co-association matrix and obtaining the final matrix with enhanced clustering performance. Experimental results on ten datasets demonstrate that the proposed approach significantly outperforms ten existing ensemble clustering approaches. Specifically, the proposed approach achieves an average improvement of 8.5% in the normalized mutual information, 7.2% in the adjusted rand index, and 6.8% in the F-measure compared to the best-performing baseline approach.