Performance measures for stochastic Discrete Event Systems (DES) often involve finite or infinite horizon expectations of measurable costs and benefits. In its broader sense, the term ``sensitivity analysis'' refers to the estimation of the impact of changes in expected performance upon changes of some of the input parameters. In the particular case where the expected performance is differentiable, sensitivity analysis deals with the estimation of gradients of the expected performance with respect to some parameter of interest, called the control variable.

The past two decades have seen a fruitful period for developping sensitivity estimators for stochastic DES, and successful implementations include applications in telecommunications, manufacturing, finance, queueing and inventory systems. The rapid pace at which seemingly different methods have been proposed make it dificult even for experts to oversee their similarities and differences. The approach that we follow here differs somewhat from the texts available to date (see references) in that we propose a methodological study and explain how different estimation techniques can be ``recovered''. Selected examples will serve as the basis for the construction of the corresponding sensitivity estimators.

Sensitivity estimation can be divided in three categories: the pathwise analysis (covering the so-called perturbation analysis--or PA methods), the weak differentiation methodology, and perturbation methods, which include finite differences, harmonic analysis and simultaneous perturbations. This course focuses on the first two categories, opening with a review of the basic concepts of probability theory required in the sequel.