Multi-view subspace clustering is an important and hot topic in machine learning field, which aims to promote clustering results based on multi-view data, which are collected from different domains or various measurements. In this paper, we propose a novel tensor-based intrinsic subspace representation learning for multi-view clustering. Specifically, to investigate the underlying subspace representation, the rank preserving decomposition accompanied with the tensor-singular value decomposition based low-rank tensor constraint is introduced and applied on the subspace representation matrices of multiple views. The specific information of different views can be considered by the rank preserving decomposition and the high-order correlations of multi-view data are fully explored by the low-rank tensor constraint in our method. Based on the learned subspace representation, clustering results can be obtained by employing the standard spectral clustering algorithm. The objective function is efficiently optimized by utilizing the augmented Lagrangian multiplier based alternating direction minimization algorithm. Experimental results on nine real-world datasets illustrate the superiority of our method compared to several state-of-the-arts.