A better understanding of the mechanical performance and the failure modes of corroded reinforced concrete (RC) structures is crucial for implementing measures that reduce failure risks. Therefore, this paper proposes a probabilistic numerical framework to estimate the structural performance and to identify the failure modes as well as the main influencing parameters affecting the safety of corroded RC structures. This framework comprehensively combines experimental data, finite element method (FEM), and polynomial chaos expansion (PCE) modeling. First, the FEM for failure analysis is developed and verified with a 26-year-old corroded RC beam. The investigated case considers the effects of corrosion degree and bond behavior of the steel-concrete interface on the mechanical properties and failure mode of the corroded RC beam. Second, PCE surrogate models for serviceability and ultimate limit states are established by combining the sampling technique (e.g., Sobol sequences) and the validated FEM model. Finally, a global sensitivity analysis is conducted using the PCE model. Several illustrative cases are presented to analyze, in deterministic and probabilistic manners, the failure modes and the sensitivities of material properties and geometry characteristics for both serviceability and ultimate limit states. The results of this study could provide useful insights for understanding the main failure modes of RC structures under different corrosion scenarios.