Aftershocks of the 1989 Loma Prieta, California, earthquake are used to estimate site response along the San Francisco Peninsula. A total of 215 shear-wave records from 24 sources and 21 sites are used in a linear inversion for source and site response spectra. The methodology makes no assumptions about the shape of the source spectrum. However, to obtain a stable, unique inverse a Q model and geometrical spreading factor are assumed, as well as a constraint on site response that sets the site response averaged over two specific stations to 1.0. Site responses calculated by this formulation of the problem are compared with other studies in the same region that use different methodologies and / or data. The shear-wave site responses compare favorably with estimates based on an ω2-constrained source model. Comparison with coda amplification factors is not as close, but still favorable considering that the coda values were determined for nearby locations with similar geology, and not the same sites. The degree of agreement between the three methods is encouraging considering the very different assumptions and data used.