In a level-2 probabilistic safety analysis (PSA), two types of uncertainty have to be taken into account: the uncertainty related to random variation (variability) and the uncertainty related to limited knowledge (ignorance). We present a consistent treatment of these two types of uncertainty within a Bayesian framework. This framework allows us to translate both types of uncertainty in the basic parameters into branch probability distributions of the PSA accident progression event tree (APET). This, in turn, results in probability distributions for the different release categories. A generic Monte Carlo algorithm for drawing random samples from branch probability distributions is presented, offering the possibility to directly include information in terms of empirical data. To provide an illustrative example, the developed methods are applied to a specific APET question, related to the temperature-induced rupture of the reactor coolant system in case of a high pressure accident scenario. Although this paper addresses level-2 PSA, the proposed framework is presented in a general form to be applicable to other PSA problems.