Permanent ground offsets, constituting a prime dataset for constraining final fault-slip dis-tributions, may not be recovered straightforwardly by double integration of near-field accelerograms due to tilt and other distorting effects. Clearly, if a way could be found to recover permanent ground offsets from acceleration records, then static datasets would be enlarged, and thus the resolution of fault-slip inversions would be enhanced. Here, we introduce a new approach for extracting permanent offsets from near-field strong-motion accelerograms. The main advantage of the new approach with respect to previous ones is that it corrects for source time functions of any level of complexity. Its main novelty is the addition of a constraint on the slope of the ground velocity spectra at long periods. We validated the new scheme using collocated accelerograms and Global Navigation Satellite Systems (GNSS) records of the 2011 Mw 9 Tohoku-Oki earthquake. We find a good agreement between accelerogram-based and GNSS-based ground offsets over a range of 0.1–5 m. To improve the spatial coverage of permanent ground offsets associated with the 2004 Parkfield earthquake, near-field accelerograms were baseline corrected using the new scheme. Static slip inversion of the combined GNSS-based and accelerogram-based ground displacements indicates appreciable seismic moment release south of the epicen-ter, about 5 km into the Cholame section of the San Andreas fault. We conclude that the strong shaking observed to the south of the epicenter is directly related to the slip in that area and is not the result of local amplification.