【Objective】At present, certain progress has been made in the research on a Chinese Soil Taxonomy (CST)-based Soil Type Retrieval System, but the research has come across the following problems: (1) Ignorance of the impact of the coupling of diagnosis object, soil type and retrieval framework (Inference Process) on updating of the system; (2) Lack of consideration of division of the spatial structure of soil information carriers, which is unconducive to management of the soil information; and (3) Scope matching of the soil characteristics always expressed in traditional conditional nested statement and logical relationships between soil characteristics, making retrieval language cumbersome and redundant.【Method】In order to improve the above situations, this study introduced the concept of ontology, set soil geography and CST classification rules as its theoretic bases, analyzed spatial structure of soil entities and their relationship with CST objects (soil types and diagnostic objects) and eventually on such a basis, established an ontological model for relationship between soil entities and CST objects. To standardize the expression of soil characteristics, this paper defined soil attribute models, and divided soil characteristics into two categories, i.e. ordinary soil characteristics and complex soil characteristics. Moreover, this paper also defined corresponding predicate logics to express the relationships between types or models in logic and membership.【Result】This paper used Python language to realize construction of the ontology models and definition of the predicate logics, developed a type retrieval system covering all the four levels (Order, Suborder, group and subgroup), and cited the data of some single pedons representative of the Soil Series of China (Hubei volume) for test. And the test not only helped determine types of the pedons at all the four levels, but also recorded the entire retrieval processes, which facilitated analysis of the result later.【Conclusion】To compare with other existing retrieval models, this ontological model divides the carrier of soil information into four categories, i.e. Horizon, Profile, Pedon, and Polypedon, which facilitates management of the soil information and reduces complexity of the classification rules. Besides, the retrieval model separates the rules from the framework and can be characterized by high cohesion and low coupling, so that the model can better support the retrieval system in updating and expanding. The retrieval framework no longer takes soil characteristics as retrieval object, but encapsulates soil characteristics into soil entities, diagnostic objects, and soil types, thus elevating retrieval objects up to the level of category, which tallies more with human cognition. In brief, a method of constructing a retrieval model is proposed from the perspective of ontology in this paper. It can solve some problems of the existing retrieval models to a certain extent. However, the issue of semantics of soil characteristics and quantification of soil morphological characteristics remains to be a topic to work on. Therefore, to further improve the model, further efforts should be made in the next phase of the research to establish a soil characteristics semantic dictionary with complete metadata and information and introduce uncertain information as system auxiliary information.