Fault slip and strain during fault-creep episodes are continuously recorded by a dense network of creep-meters deployed along several major traces of the San Andreas fault system in central California. These data are analysed on the basis of theoretical faulting models to delineate the kinematics of the fault-creep process. The results indicate that fault creep is a failure propagation phenomenon, kinematically similar to seismic faulting, but with very low characteristic rates. The speed of creep propagation is not constant and is of the order of 10 km/day or less. The maximum slip velocity usually ranges from 0.1 to 10 $\mu $m/s. Both of these are five or more orders of magnitude smaller than the corresponding rates of seismic faulting. The slowness in particle motion can account for the ineffectiveness of the creep process in exciting observable seismic waves. However, the tectonic strain released by a creep event may be sizable. The largest event recorded so far has a rupture length of 6 km and a maximum offset of 9 mm, comparable to similar parameters of a shallow earthquake of magnitude 4.7.