<p>Dense non-aqueous phase liquids (DNAPLs) are often trapped as residuals in low-permeability media. These DNAPL residuals can serve as secondary pollution sources threatening groundwater quality. Numerical simulation techniques can model the migration of DNAPLs in low-permeability media and predict their spatial and temporal distribution. To establish a reliable DNAPL groundwater transport simulation model in low-permeability media, it is crucial to evaluate the sensitivity of the model parameters and structure accurately. This study presents a DNAPL migration model for low-permeability media, based on a two-dimensional sandbox experiment. Various sensitivity analysis methods, such as the Sobol and Morris methods, as well as local perturbation analyses, are used to evaluate the sensitivity of the model’s physical parameters and constitutive model structures. The results show that the inverse of the DNAPL-water injection pressure (<i>α</i>_nw) is the most sensitive parameter in both the background and low-permeability zones of the DNAPL migration model. <i>α</i>_nw significantly impacts the distribution of DNAPL saturation distribution and migration pathways. In low-permeability media, the residual DNAPL saturation (<i>S</i>_nr) is more sensitive to DNAPL retention characteristics than permeability (<i>k</i>). The sensitivity analysis of the model structure shows that its contribution to the output exceeds 87%, greatly exceeding the influence of the parameters. Furthermore, the model structure, including the van Genuchten–Mualem relative permeability model, shows the highest sensitivity. These findings emphasize the importance of selecting the appropriate model structure and optimizing key physical parameters to improve prediction accuracy, especially under complex conditions involving low-permeability media.</p>