Peak over threshold (POT) method is frequently used in the modeling of extreme values in offshore engineering. In this paper, the POT method is examined in terms of epistemic uncertainties in practical usage. Real observed ocean data with specific considerations to extreme events are analyzed. In particular, the procedures including the use of de-clustering and threshold in POT method are addressed. A key element in the application of probability and statistical theories is the estimation of model parameters. The performance of these estimation methods is tested in the context of epistemic uncertainty. This is done through a numerical simulation study for considering data samples having different tail behavior, sample size and noise conditions. The annual maximum method and the rth largest order statistic method in establishing the extreme value model are also included in the comparative study. Main focus is put on the critical issues and uncertainties that might be resulted in the established extreme value models.