The performance of optical filters with resonant waveguide gratings is predicted numerically, assuming random fluctuations of various design variables. Specifically, we derive stochastic models based on polynomial-chaos expansions (PCEs), by employing a stochastic collocation (SC) approach that exploits the rigorous coupled-wave analysis (RCWA) deterministic solver. The statistical moments of the filters' spectral response are obtained for transverse-electric (TE) polarization, and compared against reference results from the Monte-Carlo (MC) method.