Uncertainty is inherent in watershed modeling but it is not fully acknowledged in model applications. This review focuses on uncertainty issues related to the Soil and Water Assessment Tool (SWAT) model, which is one of the most useful tools for simulating nonpoint source (NPS) pollution processes. We considered numerous studies that addressed three types of uncertainty in detail, i.e., the model inputs, parameters, and model structure. It has been shown that rainfall data, in terms of the spatial rainfall variability and the accuracy of the measured data, play a key role in the accuracy of the SWAT model. Geographic information system inputs, including the digital elevation model, land use map, and soil type map, have also been identified as key sources of input errors. With respect to the parameter uncertainty and model structural uncertainty, it is anticipated that the complex, nonlinear structure, and numerous parameters included in the SWAT model may lead to a failure to identify parameters, as well as equifinality phenomenon. We also compared some widely used uncertainty analysis methods, such as the generalized likelihood uncertainty estimation and first-order error analysis, to provide reliable guidance for the application of the SWAT model. This study benefits a wide range of researchers, who are concerned with uncertainty issues in NPS pollution modeling, and it provides insights into the application of watershed models in the development of watershed programs.