Abstract: The China Experimental Fast Reactor (CEFR) represents a significant advancement in China’s nuclear energy sector, with its main vessel serving as the primary safety barrier enclosing the reactor core. However, the structural integrity of the main vessel’s welded components is susceptible to defects such as coolant-induced corrosion and high-temperature creep, posing potential safety risks during operation. While international efforts, including visual, ultrasonic, and eddy current inspections in reactors like France’s Phoenix-Ⅰ, Japan’s MONJU and India’s prototype fast reactor, have established in-service inspection protocols, research on nondestructive testing (NDT) for CEFR’s main vessel welds remains underdeveloped. Ultrasonic testing (UT), particularly phased array ultrasonic testing (PAUT), is recognized as a viable solution due to its penetration depth, adaptability to austenitic materials, and widespread application in nuclear facilities. Yet, traditional empirical probe selection methods often lead to inefficiencies, such as low detection accuracy and repeated trials. This study addresses the challenges in optimizing UT probe selection for CEFR’s main vessel welds by integrating simulation-based analysis. A comprehensive inspection technology analysis was first conducted on test blocks mimicking the welded structure, focusing on geometric parameters and acoustic properties. Subsequently, the CIVA simulation platform, developed by the French Atomic Energy and Alternative Energies Commission (CEA), was employed to model probe performance, including beam propagation and defect response. Key parameters such as probe frequency, aperture, and focal law configurations were systematically evaluated. The simulation results identify optimal probe parameters compliant with nuclear inspection standards, demonstrating CIVA’s capability to streamline probe selection and validate detection protocols. The findings reveal that conventional UT and PAUT probes exhibit limited detection rates for volumetric defects under standard gain settings. However, sensitivity improvements through gain compensation enhance defect characterization without compromising signal-to-noise ratios. While CIVA simulations account for material anisotropy in austenitic welds, real-world complexities—such as beam deviation, mode conversion, and scattering—introduce discrepancies between simulated and actual inspection outcomes. Consequently, post-simulation experimental validation is critical to ensuring detection reliability and minimizing false positives. In conclusion, this work establishes a simulation-driven framework for UT probe optimization in CEFR’s main vessel inspection, addressing gaps in existing methodologies. The proposed approach not only reduces economic and temporal costs associated with empirical probe selection but also provides actionable insights for NDT practitioners. Future efforts should prioritize experimental validation to reconcile simulation-observation mismatches and refine defect detection results, thereby advancing the safety and operational longevity of fast reactor systems.