Rainfall data, as an important input data, its temporal and spatial resolution directly affects the accuracy of watershed hydrological simulations. Weather radar has been used in business in China, but the uncertainty of radar rainfall data is large, so the two-source rainfall data fusion based on radar and rainfall stations has become an important method to obtain rainfall data with high resolution. In this paper, taking Duanzhuang watershed in the eastern foot of Taihang Mountain for example, based on the data of 18 rainfall stations in the basin and Shijiazhuang’s S-band radar data, the radar data are preprocessed, initially optimized (overall optimization and subsection optimization) and evaluated. Then 11 rainfalls in this basin are selected by three fusion methods for fusion and quality evaluation. The results show that the pre-processed radar rainfall data and the preliminarily optimized radar rainfall data have poor effects in rainfall spatial estimation This indicates that single-source radar data cannot be directly used to describe rainfall events. Among the three fusion algorithms, the rainfall proportional coefficient fusion method (q_k method) is the best, the optimal interpolation method is the second, and the mixed geographically weighted regression-gaussian function (MGWR-GAU) fusion algorithm is the worst under the conditions of spatial and temporal variation of rainfall and different station densities. In the case of q_k method, the correlation coefficients of the three inspection stations are increased to 0.51, 0.78 and 0.82 on the point scale, and to 0.98 on the basin area scale, and the rainfall changes smoothly in time and space.It shows this method can effectively improve the data accuracy of weather radar, and it is an important fusion method to obtain high temporal and spatial resolution rainfall data in the watershed.