An empirical model of the M2 oceanic tide has been computed from the harmonic constants of a subset of deep sea and coastal tide gauge measurements. The optimal interpolation of these data based on “inverse theory” uses a priori covariance functions deduced from a global hydrodynamical model. The inverse solution, produced with its associated error maps and samples of error spectra, is surprisingly good when compared to in situ data and to a hydrodynamical model.