Abstract: Reliability data is the basis of probabilistic safety assessment in NPPs. The parametric empirical Bayes models would underestimate the uncertainty of estimated parameter in the case with few failure data. Kass-Steffey adjustment could use Taylor series expansion to correct the posterior variance. The Kass-Steffey adjustment were derived in detail, and taking the initiating events as an example, the posterior variance and 90% credible interval are calculated. It is found that the Kass-Steffey adjustment is unimportant if there are many failure data, while it is noted that the Kass-Steffey adjustment is very important when there are few failure data, which make the posterior variance change 16%~99%, and credible interval change 4%~53%.