Phase delays at high frequencies are observed in body waves that travel in the Alaska slab, along its strike at 100--150 km depth. The delays, between 2--6 Hz energy and the direct 0.5--1 Hz arrival, are 0.5--1.5 s for P waves and 1.5--4 s for S waves. Such dispersion suggests a waveguide structure that parallels the slab, perhaps near its top. A channel that is 2--6 km thick and 2.5--5% slower than surrounding mantle can explain the observations. The thickness of the layer is comparable to that of subducted oceanic crust or somewhat thinner. The layer may be crust that is slow at these depths. The required velocity anomaly is too small to be due to a continuous layer of metastable gabbro yet too large to represent an eclogite layer. It may indicate a mixture of the two, or persistence of hydrated mineral assemblages to depth. ¿ American Geophysical Union 1996