The nature of watershed hydrological model is discussed. Models are always approximations for prototypes and a model cannot reflect all the characteristics of a prototype, especially for the complicated phenomena such as the formation of rainfall-runoff of a watershed. The role of scientific thinking for decomposition and assembly in the establishment of watershed hydrological models is explained. Substantial progress of watershed hydrological models can be made only if the breakthrough of both the theories for runoff yield and concentration and the assembly methods from basic hydrological phenomenon to watershed hydrological models are realized. The reasons of ill-posed problems in the inverse problems of watershed hydrological models are analyzed. Some shortcomings of the solution for inverse problems of watershed hydrological models by means of direct optimization methods are pointed out. Preliminary ideas on the solution for the inverse problems of watershed hydrological models through big data are proposed. With more accurate and refined hydrological observation, big data method has bright prospects in the solution for both direct and inverse problems of watershed hydrological models.