The basic problem of constructing a perturbed sample path (given a parameter perturbation) from information contained in a nominal sample path is considered. Two conditions, observability and constructability, which have to be satisfied for this to be feasible are identified. One approach for accomplishing this task is to develop an augmented system model which captures both nominal and perturbed system behaviour. For the case of systems with Markov properties, an explicit methodology is presented for constructing such models. It is also shown that by an observability transformation it is generally possible to satisfy constructability conditions allowing the performance of a perturbed discrete event system to be estimated by observing only a nominal sample path. In practice, a variety of techniques may be used to accomplish this goal, depending on issues such as parameter value availability and convergence speed of various performance sensitivity, estimates. >