Abstract: Human reliability analysis (HRA) during severe accidents is a crucial yet challenging technical aspect in level 2 probabilistic safety analysis (PSA). The variability of human psychology, the complexity of inter-team coordination, and insufficient simulation training have led to a scarcity of applicable HRA models internationally during severe accidents. In recent years, domestic nuclear power plants have been actively exploring HRA during severe accidents, but the applicability of the current models and methods is less than satisfactory. In 2022, the US Nuclear Regulatory Commission developed the integrated human event analysis system for event and condition assessment (IDHEAS-ECA) method, which was applicable to HRA in all risk-informed regulatory activities and provides guidance for related research. The IDHEAS-ECA method was studied and summarized, performance influencing factor (PIF) that need to be specifically considered in HRA during severe accidents were proposed, along with their corresponding descriptions. These included information availability and reliability, scenario familiarity, task complexity, mental fatigue, time pressure and stress, procedures/guidelines, environment and situation, systems, physical demands, etc. This method was then applied to the HRA of primary loop water injection during severe accidents at an HPR1000 nuclear power plant and compared with the SPAR-H method through analysis and discussion. The research indicates that severe accidents typically involve long time windows, and directly applying the SPAR-H method can result in significant deviations, potentially causing the human error probability during severe accidents to be much lower than that during design basis accidents and design extension conditions. The IDHEAS-ECA method better meets the application requirements of the full-scope PSA in current engineering projects and represents the latest international cognitive level. However, the IDHEAS-ECA method has several limitations in engineering applications. First, the recovery factor (Re) of the cognitive failure mode (CFM) in this methodology is set to 1 by default, which lacks specific guidance and results in a certain level of conservatism in the analysis results. In engineering applications, a combination of field interviews and simulator-based tracking analysis can be used to determine the likelihood of task recovery for a particular CFM, thereby establishing the value of Re. Second, the total human error probability distribution and parameters in this methodology are still under investigation, which is a common issue faced by various HRA methods. It is recommended to refer to general HRA engineering practices, provide error factors based on log-normal distribution expert judgment, or employ the constrained non-informative (CNI) prior distribution for uncertainty analysis. Third, while the method offers a wide range of PIF and attribute selections, it is prone to subjectivity. Therefore, when applying the method to projects, it is necessary to develop implementation rules based on the project’s specific circumstances to enhance the consistency of analysis results. Fourth, most of the basic human error data in this method originate from simulator and experimental data in countries such as the US and South Korea, which may limit its applicability to domestic projects. It is suggested that independent human reliability data be collected in China and the data in this method be updated. This study has significant implications for nuclear power engineering projects and the field of nuclear safety.