Most of the existing multi-view clustering methods are based on the assumption that the data is complete. However, real-world data collection faces various problems. For example, the damage of data storage equipment, limited to the current technology lead to the lack of collection means and so on. Although some incomplete multi-view clustering methods have been proposed, the existing methods mainly have the following two problems: (1) they focus on the consistency of views, but cannot fully mine the consistency of views, and only use the semantic level consistency information, and do not explore the instance-level consistency and complementarity. (2) Self-supervised contrastive learning algorithm mistakenly identifies negative and positive samples, which leads to false negative and false positive samples. These samples have a negative impact on the performance of the model. In order to solve the above problems, this paper proposes Affinity Matrix Guided Multiple Contrastive Learning Incomplete Multi-view Clustering (AMIMVC). First, we use a high-order random walk to construct a kinship matrix, and use the affinity matrix to guide the contrastive learning of the same view and different views at the instance level, the complementarity of views is utilized, and the low-dimensional consistency is mined. The guidance of the affinity matrix in contrastive learning alleviated the problem of false negative and false positive samples recognition. After a large number of experiments, it is confirmed that our proposed method has good performance on both complete views and missing views.