We explore the method of incremental learning for an existent hierarchical latent class model which are widely used for cluster analysis of categorical data. As new data is observed, we can not ignore the new information in the data, hence it's important to improve the performance and accuracy of the model. Previous works pay little attention on incremental learning method about incomplete data, in this paper, we introduce a new approach that can sequentially update the hierarchical latent class model when new data is available, and the data coincidence degree is defined to evaluate the latent nodes that are influenced by the new data. A learning algorithm is developed and we also present experiment that demonstrates the feasibility of our approach.