Nuclear safety studies, based on the so-called BEPU (Best Estimate Plus Uncertainty) approaches, aim to calculate not only the possible values of a physical variable of interest, but also to quantify its associated uncertainty. From the results of a BEPU study, statistical analysis tools aim to improve the understanding of the physical phenomena simulated by the computer codes. The data outputs generated by these codes typically possess a functional nature, i.e. they represent the temporal evolution of a physical parameter throughout a transient. However, this functional nature is not always taken into account, in spite of the fact that it may provide relevant information regarding nuclear safety. On top of that, the functional analysis of data is even more relevant for transients where the safety criteria is directly associated to the dynamic behavior of a physical parameter, as it is in the case of the pressurized thermal shock. This work addresses the automatic identification of atypical transients (called “outliers”) in sets of time-dependent simulations that can help to better detect the physical phenomena that influence the safety margins, to find penalizing scenarios, or to verify the physical consistency of industrial simulators. A new functional outlier detection technique is then presented, as well as the eventual statistical link between the outlying simulations and the inputs of the computer code. The relevance of this methodology is illustrated on pressurized thermal shock simulations.