As the agricultural data has such complex characteristics as seasonality, timeliness, multi-level nature as well as novelty, furthermore, the present data mining software is closely coupled, data mining algorithms is hard to reused and inflexible replaced, this paper presents a theoretic framework for Web-based agricultural data mining platform. Then, it implements prototyping platform called agri-Miner with Responsibility Chain Pattern, thus to ensure this architecture to be scalable and data mining arithmetic to be replaced. Finally, this platform is successfully applied in data analysis as plant diseases and insect pests, so to verify its effectiveness of Web-based agricultural data mining platform.