Interest in the use of vadose zone transport models for fumigant risk assessment is increasing. Good modeling practice includes an assessment of model sensitivity and output uncertainty. This computational study evaluated the sensitivity of HYDRUS-1D- and HYDRUS 2D/3D-simulated fumigant cumulative flux, maximum 6-h period mean flux density, and soil gas concentrations to 15 model input variables using Monte Carlo Latin hypercube analysis. The input variables included fumigant physicochemical properties, agricultural film (tarp) properties, and soil properties. Three different application scenarios were investigated: tarped broadcast, tarped bedded shank injection, and a tarped drip line-source application. Model sensitivity to initial water content (θi), saturated water content (θs), and tarp permeability varied among scenarios depending on the relative importance of soil gas diffusive resistance and tarp mass transfer resistance to fumigant volatilization. Model outputs were sensitive to fumigant air–water partition and gas-phase diffusion coefficients, two parameters that probably have a small contribution to overall modeling uncertainty because accurate estimation methods for these parameters are available. Sensitivities to the fumigant degradation rate were high in all scenarios, and sensitivity to tarp permeability was high only when substantial volatilization occurred from the tarped portion of soil surfaces. Existing literature data for both degradation and tarp permeability are highly variable; parameterizing these processes using literature estimates may contribute substantially to model uncertainty. In several cases, the highest output sensitivities were to θs. For model comparisons to site-specific field data, soil texture-based estimates of θs are potentially large contributors to model uncertainty; direct measurement is recommended.