Spectral inversion of Lg from earthquake sources is subject to significant errors, due to random and systematic deviation of the Lg spectra from idealized theoretical models. I present an analysis of these errors and demonstrate that their effects tend to increase with decreasing frequency and recording distance. I propose modifications to the inverse method of Xie (1993) to reduce these errors. The modifications include a more robust Q estimator and the added capability of incorporating a priori knowledge into the inversion. The modified method is applied to Lg spectra from the 1995, western Texas earthquake (MW = 5.7). I measured average two-station Q values for selected station pairs, as well as the source rise time of the mainshock, with an empirical Green's function approach. These served as a priori knowledge for the simultaneous inversion for Lg seismic moment (M0), corner frequency (fc), and path Q. The result of the inversion indicates that Lg source radiation is nearly isotropic. The M0 and fc values estimated using Lg from the mainshock and two aftershocks are typical for Lg from central and eastern North American earthquakes, with M0 scaling with fc−4, and M0 and fc being 3.3 × 1016 N-m and 0.3 Hz, respectively, for the mainshock. The 1-Hz Lg Q correlates with tectonic environments, increasing from about 200 for westerly paths to above 700 for easterly paths. Two-station Q measurements suggest that, as previously discovered for Eurasia, the frequency dependence of Lg Q may be variable with distance if the distance spans a wide range (103 km).