The existing incomplete multi-view clustering methods have some problems, such as failing to make full use of the potential information of missing views, and failing to make full use of the complementary information and high order correlation among views. In this paper, a new incomplete multi-view clustering method based on multi-level self representation constraints (CMLC) is proposed. The CMLC uses the common latent representation to recover the missing value and thus effectively obtain the latent information of the missing part. In order to obtain a unified low-rank representation of multi-view data, the CMLC first captures the consistent information within the multi-view data and the complementary information between views through multi-level self-representation constraints. At the same time, it uses multi-level error representation to improve the robustness of the model against noise, and then captures the higher-order similar information between views through the logarithmic tensor. Finally, the distance regular term is introduced to capture the local information of the data. The results show that the CMLC has the best clustering performance compared with nine methods on six imperfectly simulated multi-view datasets with different miss rates.