Whether deep earthquakes beneath the southern East African Rift, down to 44 ±4 km, occur above or below the Moho has remained controversial. We explore a new technique, relying on the amplitude ratios of Sn and Lg seismic waves, to test earthquake depths relative to the Moho. Sn and Lg waves propagate through reflections in the mantle lid and the crust, respectively, so Sn is only strongly excited by events with hypocenters below the Moho and Lg by events above the Moho (see presentations in S024: Wang and Klemperer, and Chen et al., this meeting). We first validated our Sn/Lg method with an earthquake accepted to be below the Moho in the offshore extension of the East African Rift System, a magnitude 4.6 event at 14.7 3.4 km in the Mesozoic oceanic crust of the Mozambique Channel. Even given the Miocene-to-Recent volcanism (associated with the offshore extension of the East African Rift System), this event - if its depth is correct - is almost certainly in the mantle. As predicted from our Sn/Lg models, regional seismograms for this event have strong Sn but essentially no Lg. We then applied our method to the 1998 m=4.7 earthquake beneath southern Lake Malawi to determine its position relative to the Moho. Yang & Chen (JGR, 2010) report it to be 44 km deep in a region where crustal thickness varies rapidly from ~35 km within the rift to ~45km beneath the adjacent craton. The published location beneath Lake Malawi suggests the earthquake may have occurred in the mantle but Craig et al. (GJI, 2011) note that it lies within error of the Moho. Our analysis shows rather small Sn but strong Lg energy, so we believe it is likely in the lower crust. Given our results, either the current depth is incorrect or the event is not beneath the axis of the rift. Neither solution easily explains the phases Yang & Chen described as underside PmP reflections, which were used to suggest this earthquake occurred below the Moho. A possible answer may lie in strong 3D structure around the hypocenter, which may either be affecting our Sn and Lg observations, or may offer a multi-pathing solution to the delayed “underside PmP” phases