A new submodule clustering method via sparse and low-rank representation for multi-way data is proposed in this paper. Instead of reshaping multi-way data into vectors, this method maintains their natural orders to preserve data intrinsic structures, e.g., image data kept as matrices. To implement clustering, the multi-way data, viewed as tensors, are represented by the proposed tensor sparse and low-rank model to obtain its submodule representation, called a free module, which is finally used for spectral clustering. The proposed method extends the conventional subspace clustering method based on sparse and low-rank representation to multi-way data submodule clustering by combining t-product operator. The new method is tested on several public datasets, including synthetical data, video sequences and toy images. The experiments show that the new method outperforms the state-of-the-art methods, such as Sparse Subspace Clustering (SSC), Low-Rank Representation (LRR), Ordered Subspace Clustering (OSC), Robust Latent Low Rank Representation (RobustLatLRR) and Sparse Submodule Clustering method (SSmC).