Conceptual clustering analysis is suitable to discover the knowledge in database with incomplete or absent domain background information. It is difficult for original conceptual clustering method to deal with the data objects described by numerical attribute values. A new criterion function based on semantic distance is proposed in this paper, and a novel domain-based dynamic conceptual clustering algorithm (DDCA) is also presented. With the discretization of the continuous attribute values, it works well on the datasets that are described by mixed numerical attributes and categorical attributes. The algorithm automatically determines the number of clusters, modifies the demoid according to the frequency of the attribute values within each cluster and gives out the interpretations of the clustering with the conceptual complex expression. The experiments demonstrate that the semantic-based criterion function and the dynamic conceptual clustering algorithm are effective and efficient.