Up to now, the optimization of surveillance test intervals (STIs) is performed at the system level. In other words, the STI of a system is optimized considering only the conditions related to that system. For instance, the STI of an emergency diesel generator (EDG) is determined considering only the availability of an EDG and the costs related to the changed STI. However, such an approach can cause problems when the effects of each system's optimized STI are combined. That is, the core damage frequency can increase to a level that cannot be accepted by the regulatory body when the STIs optimized at the system level are all adopted together. In this paper, STIs of the systems are optimized at the plant level based on the simplified probabilistic safety assessment (PSA) model of a pressurized water reactor. The PSA model includes most of the important safety systems. It is a nonlinear and multimodal optimization problem with constraints that it optimizes the STIs of various systems based on the PSA model at the plant level. Most conventional optimization techniques have difficulties in handling such multimodal and nonlinear optimization problems. Therefore, we applied a genetic algorithm to the optimization of STIs. The genetic algorithms guarantee the global optimum and find the solution very effectively. In addition, the fault trees used in PSA have some limitations in representing the real world; i.e., in estimating the unavailability of standby systems and the effects of maintenance strategies. So, the analytical unavailability model is implemented to overcome such limits of the conventional fault tree approach. The analytical unavailability model enables us to accurately estimate the effect of a maintenance strategy on the unavailability of systems. The optimized STIs based on the conventional fault tree and the analytical unavailability model are compared.