Distribution with depth of mantle anisotropy is uncertain. To better constrain this distribution, we analyze records of the mantle Ps phases at the GRF array in Germany. Two components of these phases (SV and T) are detected by stacking teleseismic records of events well distributed in azimuth and distance. The detected T component signals are generated not so much by splitting of SV as by conversion from P to SH at the mantle discontinuities. To reconcile these observations with previously known observations of splittings in SKS, at least three anisotropic layers are required. The data favor a model with NNE oriented fast velocity at the top of the mantle. This is consistent with the results of other seismic studies. Beneath this layer the fast direction of anisotropy changes by around 90¿. Our analysis suggests as well that a layer at the base of the mantle transition zone (MTZ) is anisotropic. This anisotropy might arise in the convective boundary layer between the upper and the lower mantle. ¿ American Geophysical Union 1996