Statistical analysis has been playing an important role in the progress of society and the development of science and technology. Numerical analysis, data processing and knowledge mining are the main content of the ongoing project "Digital Forestry" run by the State Forestry Administrative, which are all dependent on statistical analytical methods and software. So it is imperative to design and develop such software related to forestry. With the support of "948" project, our research group undertake the task of designing and developing the FORSTAT system.After summarizing the technical basis of statistical software, we determined the project orientation of the FORSTAT system and analyzed the key technology of its kernel. The article consisted of six parts as follows:1. From the view of software project, the overall framework structure of FORSTAT was established by applying the Client/Server three-layer software structure and the component programming design. Reasonable function classification was made for medial logical layer through components. The transfer relationships and levels among components were identified, which showed the rules of information flow in the system. Different loading forms were designed for components with different functions, which can uninstall unnecessary components in time during system operation and recycle memory resources.2. With non-linear regression as an example, the designed model for providing components with special functions was developed with a series of procedures of initial function requirements, logical design (component structure design and user-selected interface design included), interface design, algorithm design and code design. These are significant for the upgrade of current IBM-PC procedures, function extension and system maintenance.3. A special real-time data exchange model was proposed and established, which can explain multiple relationships between dependent variables and independent variables in developing model systems (measurement error models) in forestry and ecology fields through vectored expression evaluation. Measurement error models based on the model was developed for the first time in domestic and foreign statistical software.4. Large data processing was relatively well solved by means of compiling and running of expression evaluation and code optimization. The results showed that it was efficient for frequently repeated calculations such as multiple itinerations of non-linear regression, net search for the best initial values, the parameters' partial derivative to model expression etc. It had obvious advantages in basic data operation such as copy and paste etc, especially in the condition of large data.5. Error handling mechanisms were unified for various programming languages and component development process in the FORSTAT menu-driven part, which provided detailed error information (error types, the program segments where error occurs and error descriptions included) to facilitate the user to correct the error timely.6. Data types, control flow and procedure of FORSTAT language were designed based on normal rules of procedure language design and orientation of FORSTAT language. Morphology, syntax and semantics of FORSTAT language were put forward using BNF from the aspects of machine translation, compiled executing of the language. Partial functions of the language such as compiler at statement were completed in the fields of data memory models, procedure environment allocation. Users can not only perform all menu-driven functions, but also design new algorithm themselves by second development with FORSTAT language, organize batch processing easily, and register computational process.