We compare the predictions of an a priori model of the upper mantle with seismic observations of surface waves and eigenmodes. The 3-Dimensional Seismological Model A Priori Constrained (3SMAC) has been developed by Nataf and Ricard <1996>. It is based on the interpretation by geodynamicists of the near surface layers of the Earth; on distributions of temperature, pressure, and composition as a function of depth; and then on estimates of seismic parameters (density, velocities, attenuations) from solid state laboratory measurements as a function of temperature and pressure. The 3SMAC predictions are confronted with observations consisting of phase velocities for Love and Rayleigh waves in the period range of 70--250 s <Montagner and Tanimoto, 1990>. We first show that tomographic inversions applied to 3SMAC synthetics induce a strong smoothing of the heterogeneities. This casts doubt on the meaning of the spectra of mantle heterogeneities revealed by tomography. We then show that most of the Love and Rayleigh fundamental mode observations for periods less than 200 s are satisfactorily predicted by 3SMAC. The major differences come from the seismic velocities under the Red Sea and Southeast China, which are much slower than what is estimated from 3SMAC, as well as those under Greenland, which are not as fast as the other cratonic areas. Because the lithosphere is thinner than 100 km under oceans and thinner than 300 km under continents in 3SMAC, we suggest that the existence of deeper lithospheric anomalies as proposed in many tomographic models is mostly due to a spurious effect of the inversion rather than implied by surface wave data. Half of the variance of the degree 2 anomaly mapped by low-degree eigenmode observations can be explained by lithospheric velocity structures. The other half is highly correlated with the distribution of deep slabs, but its amplitude is a factor of 3 or 4 larger than that predicted by 3SMAC. The lithospheric anomalies present a degree 6 pattern well correlated with the distribution of hotspots even when the thermal anomalies that could be associated with plumes are not included in 3SMAC. Our results emphasize the importance of giving very close attention to ''surface corrections'' in tomographic models. ¿ American Geophysical Union 1996