Data from HGLP instruments at KIP and MAT and WWSS LP and SP instruments at KIP and GUA have been used to study the amplitude characteristics of ScS and multiple ScS waves from deep-focus earthquakes. The data at low frequencies (0.006 to 0.06 Hz) are consistent with our previously published estimate, QScS = 156 ± 13 (Jordan and Sipkin, 1977). However, at high frequencies (>0.1 Hz), QScS appears to increase rapidly with frequency. Lower bounds on QScS are obtained by assuming a flat source spectrum and ignoring any energy losses due to scattering; we find that QScS must be greater than 400 at frequencies between 1 and 2.5 Hz. Correcting for a source spectrum with a corner at 0.16 Hz and an asymptotic roll-off of ω−2, considered appropriate for these events, raises this estimate to about 750. The increase in QScS at frequencies above 0.1 Hz is consistent with a spectrum of strain retardation times which has a high-frequency cutoff in the range 0.2 to 1.0 sec. At very low frequencies QScS can be estimated from normal mode data; the best available models yield values of about 230. Comparison of these estimates with our data suggests that QScS decreases with frequency in the vicinity of 0.01 Hz. Because the scattering coefficient increases rapidly with frequency, the fact that significant ScS amplitudes are observed at high frequencies implies that any bias in QScS measurements due to scattering at low frequencies is probably small. We show that, although our data provide only integral constraints on the variation of Qμ with depth, the regions in which Qμ is frequency dependent occupy a substantial portion of the mantle, probably including at least part of the mantle below 600 km depth.