Environmental models typically possess large uncertainty due to contributions from model structure, assumptions, parameterization, and data errors, not to mention lack of consideration of problem framing and the associated choice and justification of objective function. Sensitivity analysis (SA) is a fundamental tools to help identify uncertainties relevant to the modelling objectives. It provides information on the impact on model outputs from inputs and parameters and can contribute to simplifying models to make them more identifiable. However, sensitivity analysis can produce different results in accordance with several sources, such as input forcing, objective function, and the sampling undertaken. The Sobol’ method of SA is applied here to the Soil and Water Assessment Tool (SWAT). The method is based on variance decomposition, is categorized as a global sensitivity analysis (GSA) and is known to be model independent. It is able to handle non-linearity and non-monotonic functions and models. This study illustrates its findings using the total sensitivity index which includes the main effect and parameter interactions. Quasi Monte Carlo is invoked as the sampling method. The SWAT model can be regarded as an example of a complex, dynamic, over-parameterized environmental model, albeit in the hydrology domain. It has been used to simulate both water quantity and quality. It represents catchment processes based on spatially distributed soil type, weather variables, topography and land use. Multiple modules in SWAT handle hydrologic processes, weather conditions, erosion and nutrient processes. The SWAT model used here is based on previous studies by Leta et al. (2015) and Zadeh et al. (2015) for the Senne river basin in Belgium. This paper investigates how individual sources affect the results of a Sobol’ global sensitivity analysis of the SWAT model. Sensitivity analyses are performed with different weather conditions and multiple objective functions, and the stability of the ranking of parameter sensitivity is discussed. The objective functions used as illustration in this study are: the Nash-Sutcliffe Efficiency (NSE), the modified NSE (NSE*), NSE*Log, and NSE*combined. The study analyses the sensitivity indices and rank of the parameters for different weather conditions, using wet and dry calendar years selected from the five-year observation period. In case of the selected wet year, NSE and NSE* produce the same rank for parameter sensitivity. The objective functions NSE*Log and NSE*combined both return different sensitivity indices and rankings to NSE and NSE*, as they emphasize low flows and mid flows more than high flows. The SWAT parameter Cn2 (runoff curve number) becomes more influential in drier conditions whereas Ch_K2 (effective hydraulic conductivity), for example, yields lower sensitivity indices for the dry year. In addition, the study presents a visual comparison of the stability of relative sensitivities with the different sources using the estimated confidence intervals for different numbers of sampling runs. The SWAT model is generally insensitive to most parameters indicating that some of these parameters may require other conditions (i.e. a different catchment/climate) in order to be calibrated. This emphasises the need for GSA to determine which parameters are important for a given catchment when using very heavily parameterized models.