Reliable use of regional seismic phases for discrimination and magnitude estimation requires accurate corrections. Procedures that simultaneously invert for source, attenuation (Q), spreading, and site parameters have trade‐offs that result in large errors for source and distance corrections. This motivates our efforts to improve corrections by constraining trade‐offs. Using an empirical Green’s function approach, relative spectra of regional phases are computed for nearby, similar earthquake pairs of different moments, to cancel path, site, and focal mechanism effects, giving reliable estimates of source corner frequencies and relative moments. Many such pairs are available for this analysis throughout Eurasia. A large dataset of three‐component regional seismograms from Incorporated Research Institutions for Seismology (IRIS) is assembled and processed for events listed in the preliminary determination of epicenters from 1989 to 2009. A relative Brune (1970) source model is fit to network‐median relative spectra for over 46,000 pairs, corresponding to about 9400 unique events. Pseudorelative spectra are also computed from coda envelopes in 16 frequency bands. Coda is less sensitive to focal mechanism, event separation, and station coverage (Mayeda et al., 2007) but more prone to data quality issues. Results are presented with good corroboration of the moments and corner frequencies from coda and direct phases. Detailed case studies are shown to indicate the level of agreement, interstation variability, comparisons to published results based on local networks, and causes of various discrepancies between coda and direct phases. The spectra subsequently are corrected for source terms to estimate more reliable Q, geometric spreading rates, and site effects. Examples are compared to amplitude tomography results. Our estimated spreading rates are consistent with published studies, except for long distance Pn and mantle P. As important, the spreading analysis also provides a very consistent set of absolute scalar moments. Further details and comparisons to independent Q and frequency‐dependent site terms are presented in a companion paper (Fisk and Phillips, 2013).