The frequency of loss of offsite powerv(LOOP) initiating event in nuclear power plants (NPPs) in China was estimated by parametric empirical bayes (PEB) method in this paper. The prior distribution (overall distribution) of this initiating event frequency and the posterior distribution of each unit were obtained. Comparison with other commonly used methods-maximum likelihood estimate (MLE), Jeffreys noninformative prior shows that: PEB method can handle different data sources to form a larger sample. In the process of generating a prior distribution, PEB method can reflect the differences between different data sources. Therefore, it can be considered that PEB method has significant advantages in general data processing.