An unusually complete set of measurements (including rapid rate GPS over the first 10 days) of postseismic deformation is available at 12 continuous GPS stations located close to the epicenter of the 2004 M6.0 Parkfield earthquake. The principal component modes for the relaxation of the ensemble of those 12 GPS stations were determined. The first mode alone furnishes an adequate approximation to the data. Thus, the relaxation at all stations can be represented by the product of a common temporal function and distinct amplitudes for each component (north or east) of relaxation at each station. The distribution in space of the amplitudes indicates that the relaxation is dominantly strike slip. The temporal function, which spans times from about 5 min to 900 days postearthquake, can be fit by a superposition of three creep terms, each of the form ??l loge(1 + t/??l), with characteristic times ??, = 4.06, 0.11, and 0.0001 days. It seems likely that what is actually involved is a broad spectrum of characteristic times, the individual components of which arise from afterslip on different fault patches. Perfettini and Avouac (2004) have shown that an individual creep term can be explained by the spring-slider model with rate-dependent (no state variable) friction. The observed temporal function can also be explained using a single spring-slider model (i.e., single fault patch) that includes rate-and-state-dependent friction, a single-state variable, and either of the two commonly used (aging and slip) state evolution laws. In the latter fits, the rate-and-state friction parameter b is negative.