Most existing multiobjective Genetic Algorithms, like NSGA-II, employ various kinds of deterministic ranking schemes to select and evolve non-dominated solutions over the entire objective space. Such algorithms, however, perform poorly when applied to real world problems that contain noise and uncertainty in their objectives. In this paper, we propose a clustering based modification to the NSGA-II ranking scheme that improves the performance of the algorithm in noisy environments. The scheme is tested on some benchmark problems where noise manifests itself in the fitness function.