Seismic waveform data recorded at 8 to 22 km distance from about 50 buried nuclear tests detonated at the French Centre d'Expérimentation du Pacifique were analyzed in order to retrieve the isotropic seismic moment rate signature of the source. A horizontally layered seismic propagation model was considered, and only two vertical components of data were used. Taking advantage of specially designed time windowing and spectral smoothing techniques, unbiased estimates of the reduced displacement potential (RDP) were obtained in the frequency domain with the moment tensor inversion method. Using a mean regression relation calibrated from known radiochemical yields, apparent seismic yields could then be estimated from the RDP within an uncertainty of 60% at the 95% confidence level. A posteriori corrections for source-coupling efficiency and local site propagation effects were calculated empirically. For saturated volcanic hard rock, source-coupling efficiency decreases when the mean density of the source environment increases. Propagation effects are strongly correlated with the 3D geological structure of the test site and are responsible for time delays, large backazimuth anomalies, and strong variations in seismic coupling. Correcting the apparent seismic yields for a priori estimated source-coupling and propagation effects is demonstrated to reduce the yield estimation errors to less than 30% at the 95% confidence level.