Abstract: The lower head molten pool model is an important model for the in-vessel retention (IVR) effectiveness evaluation. It has the characteristics of complex relationship, many input parameters and large uncertainty. The traditional local sensitivity analysis method has the disadvantages of large amount of calculation and low efficiency when performing sensitivity analysis of complex models. In the paper, based on variance decomposition of the global sensitivity analysis method, the sensitivity analysis library (SALib) and IVR analysis code CISER developed by Nuclear Power Institute of China were adopted to analyze the sensitivity of the input parameters for five key parameters, such as the lower head wall heat flux ratio. The sensitivity coefficient and influence tread of input parameters on key results were obtained, which provides a reference for the optimization of the lower head molten pool model and severe accident mitigation strategies.