The Mw = 7.5, 1999 Chi-Chi, Taiwan earthquake occurred in the center of a dense GPS network. In the first 200 days after the earthquake displacements of as much as 10 cm accumulated in both the horizontal and vertical components. The dense spatial coverage and extraordinary signal to noise ratio make the postseismic deformation field of the Chi-Chi earthquake arguably the best ever recorded. Preseismic displacement rates provide unique information on the geometry and slip-rates of active faults, the earthquake loading cycle, and tectonics of an active arc continent collision. Afterslip, viscous flow, and poroelastic relaxation have all been proposed to explain transient postseismic deformation. The data from the Chi-Chi earthquake is of such high signal to noise ratio that it is possible to discriminate between these processes. The early postseismic transient (3 months) is best explained by afterslip, however analysis of deformation over the subsequent years may reveal viscoelastic and/or poroelastic relaxation. Inversion results reveal that afterslip encircled the zone of large coseismic slip, consistent with stable slip being driven by stress changes caused by the earthquake. Inferred afterslip at seismogenic depths raises the question of why this slip occurred slowly rather than rapidly during the earthquake. Space-time inversions combined with mechanical models of slip consistent with laboratory derived friction laws will help to address this and should also elucidate fault zone properties and stresses. Inversion of both the coseismic and postseismic GPS data are consistent with a ramp-flat geometry for the Chelungpu Fault. The geometry of active faults at greater depths is, however, not fully resolved. One difficulty has been that mechanically consistent models of interseismic deformation in compressional orogenic environments are needed. We propose to develop physical 2D viscoelastic models of the earthquake cycle in Taiwan, which include the first order effects of gravity, and in which the slip rates on faults are driven by far field plate motions rather than imposed kinematically. These models, combined with the available velocity field, will allow estimates of the geometry and slip-rates on active faults in Taiwan.