We reconsider the analysis of the 93 PASSCAL-OREGON SRP experiment data. The techniques applied to the processing of Pds waves, which are P to S conversions at seismic discontinuities at depth d in the receiver region, differ from those employed by Dueker & Sheehan [1997>. We perform global move-out corrections and migration, taking advantage of the alignment of seismicity along the great circle described by the array (South America, and the Kuril and Aleutians islands). We do not detect a deep signature of a mantle plume. The 410 km discontinuity exhibits a fragmented aspect, which prevents us from reliably estimating the transition zone thickness. The 660 km discontinuity is clearly more visible, and largely deflected in a large part of the profile. We associate this anomalous behavior and the noisy aspect of the discontinuities with the presence of a Farallon plate fragment. ¿ 1999 American Geophysical Union