The application of physics based distributed models in watershed analysis has increased in the last few decades due to increased availability special data and advanced special analysis tools. These models have the advantage of better representation of the system which produces more accurate and reliable outputs. However, they are characterized by their large parameter size, which causes lot of uncertainty in the model output. The calibration of such models is also typically complex due to the large number of parameters. Sensitivity analysis of the parameters is performed prior to calibration in order to simplify the calibration procedure; nonetheless the commonly employed sensitivity analysis methods generally do not consider the nonlinear relationship between the parameters and the output. The current study focuses on performing the nonlinear sensitivity analysis on a distributed watershed model. The method utilizes the Sobol's sensitivity approach, which is based on decomposing the total variance of the model output in terms of individual parameter's contribution. The parameter sampling is performed by Latin hypercube sampling in this study. The output of the analysis help ranking the parameters of the model in terms of their sensitivity towards the model output, thereby help pruning the parameters that are to be calibrated. The method is demonstrated using soil and water assessment tool (SWAT) developed for St. Joseph watershed, USA.