The temperature gradient of the thermal boundary layer in the D″ region is one of the basic properties of the mantle that remains difficult to evaluate. We characterize this quantity for a region in the western Pacific using gradient-sensitive seismological probes. First, both the amplitudes and travel times of the diffracted P and S phases (Pdiff, Sdiff) were analyzed to constrain the regional-scale vertical gradients of VP and VS. Acceptable models were grid-searched using synthetic waveforms for trial models that employ only one velocity gradient zone in D″. The VS models were searched over the parameter space of the thickness (H) and the gradient. Solutions show negative deviation from Preliminary Reference Earth Model and a trade-off between gradient and H from -0.00088 s-1, for190 km to -0.00049 s-1, for 240 km with a typical error of ¿0.00015 s-1. Rather than pinning down the best solution, we consider two sets of solutions for VS with H = 190 and 240 km. For these two H values, the gradients of VP were determined solely by amplitude to be -0.00032 s-1 for 190 km and 0.00015 s-1 for 240 km, shrouded in relatively large errors of 0.0002--0.0003 s-1. We derive the temperature gradient from the resolved velocity gradients using the seismological-thermodynamic equation of Doornbos et al. (1986) and arrive at 9.0 ¿ 5.8 K/km and 5.7 ¿ 4.1 K/km over 190 and 240 km, respectively. The difference in gradient between these two models is statistically significant at the 99% confidence level, and the odds that the apparent steeper gradient is caused by squeezing energy of diffracted waves into a thinner zone can also be rejected with high confidence. The combination of the two models is thus consistent with a nonlinear temperature profile that steepens toward the core-mantle boundary. Both models detect the superadiabatic gradient of temperature in the lowermost mantle.