Global sensitivity analysis offers a set of tools tailored to the impact assessment of certain assumptions on a models output. A recent book on the topic covers those issues [1]. Given the limited space for discussing thoroughly any of those methods, we will next summarize the main conclusions that derive from the application of various global sensitivity analysis methods on chemical models [2], econometric studies [3] financial models [4] and composite indicators [5,6].The incentive for subjecting a model to sensitivity analysis is linked to the aim of quantifying the relative importance of the model input factors and providing answers to specific questions, such as "Which of the uncertain input factor(s) is so non-influential that we can safely fix it/them?" or "If we could eliminate the uncertainty in one (or just few) of the input factors, which factor(s) should we choose, so as to reduce the most the variance of the output?" The choice of the proper sensitivity analysis technique depends on considerations such as: (a) computational cost of running the model, (b) number of input factors, (c) features of the model (e.g. linearity), (d) consideration of interactions among the input factors in the model, (e) setting for the analysis and audience. These characteristics may be used as criteria for deciding on the proper sensitivity analysis method to be used, although, as it is often the case, more than one method might be suitable for the same study.In case of models that require a modest amount of CPU time (i.e. up to the order of one minute per run), and with a number of input factors which does not exceed, for example, 20, the class of variance-based techniques (Sobol' method [7]) yields, convergence aside, the most complete and general pattern of sensitivity. An easy-to-code implementation [8] provides all the pairs of first-order and total effect sensitivity measures at a cost of (k + 1)N model runs, where is the number of factors and sample size N ≈ 500 ÷ 1000. The first order sensitivity measures capture the direct impact of an input factor, whilst the total effect sensitivity measures capture the direct and indirect (due to interactions) impact of an input factor. Any interaction term between two input factors can be easily computed at the additional cost of N model evaluations per sensitivity measure. The Sobol' method does not rely on any assumption about smoothness of the input-output mapping; it only relies on squareintegrability of the output variable Y . This is both an advantage and a limitation, since it implies a quite slow convergence rate of the estimator and therefore a computational cost that depends on the dimension k of the problem and on the relatively large N required for a reasonable accuracy of the sensitivity measures. In Saltelli's recipe [8], the sampling of the input factors is based on quasi-random numbers, which are sequences of multidimensional points characterized by the property of optimal space filling. In case of correlated input factors, an ad hoc computational scheme must ...