Hepatocellular carcinoma (HCC) is the most common and lethal liver malignancy. The treatment of this disease has been hampered by its heterogeneity, which severely limits progress in its personalized therapy. Therefore, it is necessary to divide highly heterogeneous HCC into molecular subtypes with similar characteristics for its clinical treatment. With the development of high-throughput technologies, integrative multi-omics data can deepen our comprehension of the biological mechanisms behind HCC pathogenesis. And it can also open new ideas for HCC stratification studies. Cluster analysis has been the main algorithm of cancer subtypes research for many years. Based on the number of input clustering algorithm omics, we summarize the current multi-omics HCC stratification methods into two major strategies: “from single-to-multi (S To M)” and “from multi-to-multi (M To M)”. Among them, the S To M strategy is to stratify HCC using different features of single omics and then combine multi-omics data to find the differential molecules among different HCC subtypes and verify the authenticity of their differences and the association with tumor biological phenomena. Feature selection is the core of the S To M strategy. Over the years, there are 3 approaches to the selection of stratified features in the S To M strategy: data distribution-based, biological features, and multi-omics approaches. Unlike S To M strategy, M To M strategy is based on the concept of systems biology and presents a landscape of the differences and associations between different omics within different subtypes. The core step of the M To M strategy is data dimensionality reduction, which puts multi-omics data into a low-dimensional stacked matrix, providing input for the subsequent cluster analysis. Generally, M To M strategy stratification algorithms can be classified into three categories: similarity-based, integration-based, and deep learning. We believe that both S To M and M To M need to pay attention to the combination of data and the applicability and practicality of related software when using it for cancer subtype analysis. In the end, we summarized the current multi-omics characteristics of HCC subtypes. We found that HCC subtypes obtained by different methods may have a common feature, which suggests that more studies are needed in the future to summarize the more representative subclasses among them.