An exact solution for the probability of failure of large complex infrastructure systems is rarely obtainable; however the probability of failure can often be bounded. An example of this type of system is the levee system in the California Bay Delta. Large levee systems often consist of many components arranged in series and parallel sub‐systems. There is the problem of defining component (or reach) length, and therefore the total number of components in the system where component length is dependent on failure mode. Methods of bounding probability of failure based on uni‐, bi‐, and tri‐modal component probability of failure are discussed. The bounds are highly sensitive to the total number of components in the system. Characterization of spatial variability using semi‐variograms is used to define component length for various failure modes. Combining the statistically defined component length with system probability of failure bounds allow for a more accurate estimate of failure probability. Demonstration of these methods and results for specific levee systems in the California Bay Delta are shown in this paper.