Coda-duration τ at 42 of the stations in the U.S. Geological Survey's central California seismic network (CALNET) for earthquakes in five source regions of central California—the Parkfield and San Juan Bautista sections of the San Andreas fault, the Sargent fault, the Coyote Lake section of the Calaveras fault, and the Livermore area—are used to obtain empirical formulas relating local magnitude ML and seismic moment M0 to τ and epicentral distance Δ. Models with log2 τ fit the data better than those assuming a log τ dependence. For 55 earthquakes with 1.1. ≦ ML ≦ 5.3, ML = 0.92 + 0.607 (±0.005)log2 τ + 0.00268(± 0.00012)Δ. These ML assume a Wood-Anderson seismograph magnification of 2800; 0.15 should be subtracted from these ML for continuity with magnitudes obtained from or calibrated against typical (magnification ∼ 2000) Wood-Anderson seismographs. For 53 earthquakes with 18.4 ≦ log M0 ≦ 22.3, log M0 = 17.97 + 0.719(± 0.0007)log2 τ + 0.00319(±0.00013)Δ. These relations provide unbiased estimates of ML for 1.5 ≲ ML ≲ 5.3 and 19 ≲ log M0 ≲ 22.3. Station corrections can significantly improve the accuracy and precision of ML and log M0 estimates, particularly if τ from a small number of stations are used. Regional variations in station corrections reflect an increase in coda duration toward the south within the CALNET.