We review our experience with the construction of models of subsurface fracturing in geothermal fields by the inversion of shear-wave splitting (SWS) observations from natural and induced seismic events recorded by local arrays of three-component digital seismometers. SWS is a phenomenon whereby shear seismic waves split into two as a result of the mechanical anisotropy created in an otherwise isotropic rock by aligned micro-fractures. The two split waves travel at different speeds, and the polarization of the faster wave is usually parallel to crack orientation. The time delay between the two split S-waves is proportional to the number of cracks per unit volume. Success in the inversion of SWS data hinges on the assumption that the observed SWS is due solely to the mechanical anisotropy induced by aligned cracks and micro-cracks in an otherwise isotropic matrix. The presence of lithologic anisotropy and/or strong heterogeneity in the reservoir rock limits the resolvability of the method. However, despite the large amount of data and diversity of geologic settings we have studied so far, the above assumption has been found to be reliable. In practice, stability and resolution in the inversion of SWS data are the issues of utmost importance since both are critically dependent on the distribution of the two SWS measured parameters (polarization and time delay) around each seismic sensor. In this paper we discuss a few lessons we have learned as to the value of SWS for geothermal exploration, its limitations and potential extensions, from nearly a decade of practice. (c) 2005 CNR. Published by Elsevier Ltd. All rights reserved.