Soil and Water Assessment Tool (SWAT) model is a useful tool for long term hydrological simulation in large scale and complicated watersheds. SWAT model has been widely used in research of hydrological cycle, soil erosion, pollution loads, and impact of climate and land use change. This model has achieved good simulation results, but the uncertainty is widespread. From the perspective of the model user, this paper discussed the main sources of uncertainty through the three important steps in using SWAT model, including input data preparation, sub-watershed delineation and data scale transformation, and model calibration. Input data with insufficient precision is one of the main uncertainty sources, owing to its inability to represent the spatial variability. Insufficient precision of sub-watershed delineation would cause over-lump of parameters during scale transformation, thus cause simulation errors. In calibration process, uncertainty is mainly induced by usage of inappropriate observation and evaluation index, and equifinality. A group of measures is suggested to reduce modeling uncertainties. Increasing resolution of input data and precision of watershed delineation to a certain threshold would reduce input uncertainty. Localization of key parameters and simulation functions based on local literature and expert knowledge can effectively reduce equifinality. And using multiple evaluation indices, automatic calibration combined with manual calibration, and calibration multi factors and multi stations can limit calibration uncertainty. Grasping the possible uncertainty sources of the main steps during the model applying, taking corresponding measures to reduce uncertainty according to the study area’s specific characteristics and research goals, is the necessary way to improve the credibility of the simulation results of SWAT model.