In this article we present and explore the source and ground-motion model uncertainty and parametric sensitivity for the 2002 update of the California probabilistic seismic hazard maps. Our approach is to implement a Monte Carlo simulation that allows for independent sampling from fault to fault in each simulation. The source-distance dependent characteristics of the uncertainty maps of seismic hazard are explained by the fundamental uncertainty patterns from four basic test cases, in which the uncertainties from one-fault and two-fault systems are studied in detail. The California coefficient of variation (cov, ratio of the standard deviation to the mean) map for peak ground acceleration (10% of exceedance in 50 years) shows lower values (0.1–0.15) along the San Andreas fault system and other class A faults than along class B faults (0.2–0.3). High cov values (0.4–0.6) are found around the Garlock, Anacapa-Dume, and Palos Verdes faults in southern California and around the Maacama fault and Cascadia subduction zone in northern California.