National Thesauri Warehouse(NTW) is an integrated service system of Chinese thesauri. Its architecture consists of three layers: data acquisition and conversion layer, storage and semantic integration layer, service and application layer. The first layer concentrates on the acquisition and specification of thesauri data as well as the conversion of thesauri data format. The second layer, storage and semantic integration layer, adopts central thesaurus integration mode, which includes three embedded layers: a layer of classification and ontology at the top, a conceptual layer in the middle, and a basic lexicon layer at the bottom. For services, NTW adopts a progressive service rationale and provides data services of different sizes and different semantic levels. Pertaining to the macrostructure, thesauri and their components are independent and isolated conceptual schemes. The macrostructures of thesauri are demonstrated by showing relationships between individual thesauri schemes. In terms of the microstructure, concepts and terms are treated as individual describing objects. Terms are used as labels of concepts. Semantic attributes are described on conceptual layer. Non-semantic attributes are described on term layer. Finally, the metadata for thesauri is designed based on Dublin Core(DC), whereas the metadata for concepts and terms are formulated based on SKOS+XL. National Thesauri Warehouse (NTW) is an integrated service system of Chinese thesauri. Its architecture consists of three layers: data acquisition and conversion layer, storage and semantic integration layer, service and application layer. The first layer concentrates on the acquisition and specification of thesauri data as well as the conversion of thesauri data format. The second layer, storage and semantic integration layer, blind central thesaurus integration mode, which includes three embedded layers: a layer of classification and ontology at the top, a conceptual layer in the middle, and a basic lexicon layer at the bottom. For services, NTW adopts a progressive service rationale and provides data services in different sizes and different semantic levels. Pertaining to the macrostructure, thesauri and their components are independent and isolated conceptual schemes. The macrostructures of thesauri are demonstrated by looking relationships between individual thesauri schemes. In t erms of the microstructure, concepts and terms are treated as individual describing objects. Terms are used as labels describing concepts. Semantic attributes are described on conceptual layers. Non-semantic attributes are described on term layer. Finally, the metadata for thesauri is designed based on Dublin Core (DC), while the metadata for concepts and terms are formulated based on SKOS + XL.