We compile and analyze a dataset of secondary microseismic P-wave spectra that were observed by North American seismic arrays. Two distinct frequency bands, 0.13–0.15Hz and 0.19–0.21Hz, with enhanced P-wave energy characterize the dataset. Cluster analysis allows to classify the spectra and to associate typical spectral shapes with geographical regions: Low frequency dominated spectra (0.13-0.15Hz) are mostly detected in shallower regions of the North Atlantic and the South Pacific, as well as along the Central and South American Pacific coast. High frequency dominated spectra (0.19-0.21Hz) are mostly detected in deeper regions of the North-Western Pacific and the South Pacific. For a selected subset of high quality sources, we compute synthetic spectra from an ocean wave hindcast. These synthetic spectra are able to reproduce amplitude and shape of the observed spectra, but only if P-wave resonance in the water column at the source site is included in the model. Our datasets therefore indicate that the spectral peaks at 0.13-0.15Hz and 0.19-0.21 Hz correspond to the first and second harmonics of P-wave resonance in the water column that occur in shallower ocean depths (<3000m) and in the deep ocean (∼5000m), respectively. This article demonstrates the important effect of water column resonance on the amplitude and frequency of P-waves that are generated by secondary microseisms, and that the amplitude of high quality sources can be predicted from ocean wave hindcasts within a factor of 0.4 − 6.