We measured motions of 54 sites in an east-west transect across northern California at 38¿--40¿ north by Global Positioning System (GPS) observations over a 4 year. We estimate the total slip rate on the San Andreas fault system to be 39.6-0.6+1.5 mm/yr (68.6% upper and lower confidence intervals from a nonlinear inversion are indicated by superscripts and subscripts). Slip rates on the individual faults are determined less precisely due to the high correlations between the estimated parameters. Our best fitting model fits the fault-parallel velocities with a mean square error of 1.04 and the following estimated fault slip rates (all in mm/yr): San Andreas 17.4-3.1+2.5, Ma'acama 13.9-2.8+4.1, and Bartlett Springs 8.2-1.9+2.1. The data are fit best by models in which the San Andreas fault is locked to 14.9-7.1+12.5 km, the Ma'acama fault locked to 13.4-4.8+7.4 km except for shallow creep in the upper 5 km, and the Bartlett Springs fault creeping at all depths. The Ma'acama fault most likely poses a significant seismic hazard, as it has a high slip rate and has accumulated a slip deficit large enough to generate a magnitude 7 earthquake. We find little evidence for contraction across Coast Ranges, except at western edge of Great Valley where 1--3 mm/yr of shortening is permitted by the data. No strain is observed within the Great Valley or Sierra Nevada except that associated with right-lateral strike slip on the San Andreas fault system. This is consistent with models of the Pacific-North America plate boundary zone in which the relative plate motion is partitioned into two domains, one strike-slip and one dominantly extensional, separated by the elastically deforming Sierra Nevada-Great Valley block. ¿ 1999 American Geophysical Union