Components of nuclear power plants do not always have historical failure data to probabilistically evaluate their reliability characteristics. To overcome this drawback, an alternative approach has been proposed by involving experts to qualitatively justify basic event likelihood occurences. However, expert judgments always involve epistemic uncertainty and this uncertainty needs to be quantified. Existing fault tree analysis quantifies uncertainty using Monte Carlo simulation, which is based on probability distributions. Since expert judgments are not described in probability distributions, Monte Carlo simulation is not appropriate for evaluating epistemic uncertainty. Therefore, a new approach needs to be developed to overcome this limitation. This study proposes a fuzzy probability algorithm to evaluate epistemic uncertainties in fault tree analysis. In the proposed algorithm, fuzzy probabilities are used to represent epistemic uncertainties of basic events, intermediate events, and the top event. To propagate and quantify epistemic uncertainty in fault tree analysis, a fuzzy multiplication rule and a fuzzy complementation rule are applied to substitute the AND Boolean and OR Boolean gates, respectively. To see the feasibility and applicability of the proposed algorithm, a case-based experiment on uncertainty evaluation of the AP1000 long term cooling system to mitigate the large break loss of coolant accident is discussed. The result shows that the best estimate probability to describe the failure of AP1000 long term cooling system generated by the proposed algorithmis 3.15×10-11 , which is very closed to the reference value of 1.11×10-11. This result confirms that the proposed algorithm offers a good alternative approach to quantify uncertainties in probabilistic safety assessment by fault tree analysis.