Considerable efforts have been taken to estimate the radioactive source term by using integrated computer models, but uncertainties exist in using these models, partly due to the fact that the results cannot be validated using real-world large-scale experiments. The purpose of this research is to identify the MAAP4 phenomenological model parameters with large influences in predicting environmental releases. The identified parameters were compared for different accident scenarios and plant configurations. Two-step screening of the influential model parameters through the Latin hypercube-rank correlation and the one-at-time method was used for the comparative analysis. From the integral analyses, it was found that the source term error propagation was much larger in the containment and in the secondary system. However, error propagation was not necessarily larger inside the containment than in the secondary system. There are common phenomenological parameters with significant influences on the source term simulations: primary depressurization, molten core-concrete interaction, and cladding oxidation models. More focused research regarding the effect of aforementioned identified phenomenological models on the source term calculations would be necessary to better predict the amount of environmental radionuclides release for the severe accidents. However, the detailed uncertainty and sensitivity analysis involving phenomenological model parameters may have to be done in a plant specific way. Some phenomenological model parameters that showed significant impact on the global source term uncertainty in one plant type did not necessarily have the same impact on another plant. A sensitivity study involving a typical PWR may not have same implications in other designs, even if a same accident scenario with similar accident phenomena is analyzed. Such sensitivity study would be much more beneficial for standardized plant designs. (C) 2018 Elsevier Ltd. All rights reserved.