Performances and reliability of NDT operations are often quantified through the determination of Probability of Detection (POD) which can then be used in lifecycle management approaches like the damage tolerance design. Recent progresses in NDT models and POD methodology using uncertainty propagation through simulation codes have demonstrated the relevance of simulation to establish POD curves. Besides obvious cost reductions, an advantage of using simulation for POD is that it potentially gives access to very large datasets, which results in reducing the width of confidence bands. On the other hand a major difficulty in this approach is often to describe the statistics of uncertain parameters, which intuitively affects the ``confidence'' one can have in the obtained POD results. In this paper we propose a new approach of ``confidence'' for POD estimation from simulations, which consists in computing a confidence band based on the level of knowledge that we have on the statistics of the uncertain parameters rather than on the number of available data. This approach is also a way to introduce ``conservativity'' in POD with simulation, which is generally appreciated in industrial practices.