We apply a kinematic finite-fault inversion scheme to P_nl displacement waveforms recorded at 14 regional stations (Δ<2°) to recover the distribution of coseismic slip for the 2004 Parkfield earthquake using both synthetic Green’s functions (SGFs) calculated for one-dimensional (1D) crustal-velocity models and empirical Green’s functions (EGFs) based on the recordings of a single M_w 5.0 aftershock. Slip is modeled on a rectangular fault subdivided into 2×2 km subfaults assuming a constant rupture velocity and a 0.5 sec rise time. A passband filter of 0.1–0.5 Hz is applied to both data and subfault responses prior to waveform inversion. The SGF inversions are performed such that the final seismic moment is consistent with the known magnitude (M_w 6.0) of the earthquake. For these runs, it is difficult to reproduce the entire P_nl waveform due to inaccuracies in the assumed crustal structure. Also, the misfit between observed and predicted vertical waveforms is similar in character for different rupture velocities, indicating that neither the rupture velocity nor the exact position of slip sources along the fault can be uniquely identified. The pattern of coseismic slip, however, compares well with independent source models derived using other data types, indicating that the SGF inversion procedure provides a general first-order estimate of the 2004 Parkfield rupture using the vertical P_nl records. The best-constrained slip model is obtained using the single-aftershock EGF approach. In this case, the waveforms are very well reproduced for both vertical and horizontal components, suggesting that the method provides a powerful tool for estimating the distribution of coseismic slip using the regional P_nl waveforms. The inferred slip model shows a localized patch of high slip (55 cm peak) near the hypocenter and a larger slip area (~50 cm peak) extending between 6 and 20 km to the northwest.