The active Chihshang fault at the boundary between the Eurasian and the Philippine Sea plates along the Longitudinal Valley in eastern Taiwan is creeping near the surface but has also produced large earthquakes at mid-crustal depth such as the 2003, Mw 6.5, Chengkung earthquake. The creep rate measured at the surface shows strong seasonal fluctuations before the Chengkung earthquake, correlated with groundwater pressure variations measured at nearby wells. The Chengkung earthquake did not rupture the fault near the surface but induced a sudden increase of creep rate that decayed with time during the postseismic period. These observations suggest that the near surface fault obeys a rate-strengthening friction law. We conduct numerical simulations based upon the observed variations of creep rate with regard to a velocity-strengthening friction law and 1-D groundwater diffusion model to investigate the fault rheology behind this phenomenon. The model, which assumes a creeping fault segment extending from the surface to a depth of 5 km, yields a good fit to the creep data when the friction parameters are assumed to vary with depth or to have changed at the time of the Chengkung earthquake. Our best model suggests that the creeping zone is characterized by a rate parameter a=∂μ/∂log (V) of 1.3 × 10−2 and a friction coefficient of 0.84, and a long-term slip rate of 25.9 mm yr−1 at depths less than 87 m. By contrast, the lower segment of the creeping zone exhibits a much lower friction coefficient of 0.19, a smaller a of 6.6 × 10−3 and a higher long-term slip rate of 38.1 mm yr−1 at depths between 0.087 and 5.0 km. This change in rate parameters implies that the lithologies or physical properties of the fault rocks may vary with depth. Alternatively, a model assuming a dramatic increase in rate parameter from 5.6 × 10−4 to 6.2 × 10−3 by the strong ground shaking of the Chengkung earthquake can also yield a fair agreement with the observed data. The lack of a notable deceleration in fault creep rate during the dry period from 2002 to mid-2003 suggests that the real recharging system for the fault zone may be more complicated than what we assumed in this study.