We analyze several thousand high-quality, globally recorded SS-S differential waveforms to constrain the lateral variation of shear wave attenuation (Qβ) in the upper mantle. We use a multitaper frequency domain technique to measure attenuation, parameterized by a t* operator, and implement a robust estimation technique to compute t* and its variance. The differential waveform technique minimizes the effect of factors such as finite source duration and structural complexity near the source and receiver so the differential SS-S waveforms are mainly sensitive to the shear attenuation in the upper mantle under the SS bounce point. We use seismograms recorded at ranges of 45¿ to 100¿ and compute the SS-S differential t* from the broadening of the SS waveform relative to the Hilbert transform of the S waveform. A careful choice of fitting windows allows us to reduce the biasing effects of interfering phases which can affect t* by up to 0.5 s. The t* residuals (with respect to preliminary reference Earth model (PREM)) vary by ¿1.5 s with an average of ≈0.24 s. Our study suggests an average Qβ value of 112 (most of the lateral variations of Qβ are within 30% of this value) in the top 400 km of the mantle, slightly lower than the PREM value of 128. There is a qualitative correlation of t* residual with tectonic region with distinctly higher attenuation observed under young oceans compared to platforms and shields. Also, the lateral variations of the residuals are similar in trend to those observed in studies of the attenuation of ScS multiples. At long wavelengths, the Qβ map shows a modest correlation with shear wave attenuation maps computed from surface wave analyses and with the patterns of lateral variations of shear velocities at certain upper-mantle depths predicted by the model S16B30. The correlation with the velocity model is highest at 300--500 km depth indicating that there may be a contribution to long-wavelength attenuation from relatively deep regions. Formal inversion for an upper mantle Qβ model shows that while lateral resolution is quite good, depth resolution is poor as might be expected. Better depth resolution must await combined body wave and surface wave inversions. ¿ American Geophysical Union 1996