The paper presents the application of the probabilistic risk assessmen
t (PRA) technique for surveillance test interval (STI) evaluation. Ass
essment was performed for the reactor protection system (RPS) and base
d upon the living PRA model of the Taipower's Maanshan power plant. Fa
ult tree was developed and analyzed to find the most proper subsystem
for test interval relaxation. Comparisons were made for generic data a
nd plant specific data sets; approaches of staggered test versus seque
ntial test were also compared. Sensitivity studies were made to evalua
te the uncertainty of human error and common cause failure factors. Ev
ent tree analyses were made to calculate core melt down frequencies fo
r test induced transients and anticipated transient without scram (ATW
S) events and to evaluate the effect of STI relaxation. Results indica
ted that an extension of three times of the current RPS test interval
has no significant risk impact for the Maanshan plant.