Many problems of practical importance can be translated into the study of a real-valued objective function of several continuous variables. The graph of such an objective function defines a surface that we term “landscape”. Different aspects of the same landscape might be of interest, but attainable only through independent means. For example, when the main objective is optimization, one can use Evolutionary Algorithms (EAs) to generate a search path designed to be biased toward the set of optima of the landscape. If the main purpose is to compute the global sensitivity of the objective function with respect to its parameters, sampling strategies tailored to this problem are available in order to avoid a bias in the estimate due to bad sampling of the landscape. Nevertheless, when evaluations of the objective function are costly, the ability to achieve such conflicting tasks at once becomes crucial. We present here a generic estimation procedure for off-line global and local sensitivity indices from samples generated by an EA optimizer.