This paper introduces an evolutionary approach to the variability analysis of nonlinear systems. The algorithm shows some advantages with respect to classical stochastic approaches and, in particular, with respect to Monte Carlo-based methods. First, it allows a significant reduction of the underestimation error that affects any nondeterministic approach; in other words, it guarantees a considerable saving of simulation time. Moreover, the technique ensures a higher reliability in pursuing the true range of variation of the system performance of interest. The method has been applied to a practical problem of great interest: the variability analysis of a stress control tube made of a composite material used for electric stress relief in cable accessories. The cable termination has been modeled by means of a lumped nonlinear circuit. The proposed technique also provides reliable results whenever large parameter variations need to be taken into account and/or nonlinear functions of varying parameters are considered. Simulation results prove it to be of great support for the designer in cable termination dimensioning and in composite material selection.