In the design of large-scale and complex mechanical systems, determination of design parameters is a very difficult problem. This study deals with parameter satisfaction problems of large-scale, complex, and dynamic systems with judgment functions. In order to solve these problems, a new method is proposed which sequentially exchanges the original mathematical model to an analyzable approximate model by means of the identification method and which improves a lot of parameters simultaneously. First, criteria of selection of attributes to build up approximate model are clarified. Moreover, as tools for selection of attributes, two analysis charts are proposed which express dynamic relationships among the attributes. Second, a general condition to determine the structure of approximate models and a statistical method to linearize the original model with judgment functions are derived. Finally, a combinatorial method of statistical identification and parameter optimization methods are proposed. By this method, some shortcomings of sensitivity analysis, decomposition technique (enormous calculations of partial derivatives) and statistical methods such as Monte Carlo method (bad convergency of solutions) can be avoided. As a result of this, it becomes possible to search satisfactory parameters efficiently.